Language and modality: Effects of the use of space in the agreement system of lengua de signos española (Spanish Sign Language)

Language and modality: Effects of the use of space in the agreement system of lengua de signos española (Spanish Sign Language) LANGUAGE AND MODALITY: EFFECTS OF THE USE OF SPACE IN THE AGREEMENT SYSTEM OF LENGUA DE SIGNOS ESPAÑOLA (SPANISH SIGN LANGUAGE) ACADEMISCH PROEFSCHRIFT ter verkrijging van de graad van doctor aan de Universiteit van Amsterdam op gezag van de Rector Magnificus prof. dr. D.C. van den Boom ten overstaan van een door het College voor Promoties ingestelde commissie, in het openbaar te verdedigen in de Agnietenkapel op vrijdag 29 januari 2016, te 10:00 uur door BRENDAN DENIS NICHOLAS COSTELLO geboren te Manchester, Verenigd Koninkrijk (cc)2016)BRENDAN DENIS NICHOLAS (cc by-nc-sa 4.0) Promotiecommissie: Promotor: prof. dr. A.E. Baker Universiteit van Amsterdam Copromotores: dr. R. Pfau Universiteit van Amsterdam dr. M. A. Landa Arevalillo Universidad del País Vasco/ Euskal Herriko Unibertsitatea Overige leden: prof. dr. P.C. Hengeveld Universiteit van Amsterdam prof. dr. E.O. Aboh Universiteit van Amsterdam prof. dr. J. Quer Villanueva Universitat Pompeu Fabra prof. dr. M. Steinbach Georg-August-Universität Göttingen dr. V. Kimmelman Universiteit van Amsterdam Faculteit der Geesteswetenschappen This doctoral thesis was carried out in cotutelle between the University of Amsterdam and the University of the Basque Country. The research of this doctoral thesis received financial assistance from a doctoral scholarship from the Basque Government (ref. BFI05.149). Do m'athair, do mo mháthair, do mo dheirfiúr. I have been to hell and back. And let me tell you, it was wonderful. Louise Borgeois Table of contents ACKNOWLEDGEMENTS XI ABBREVIATIONS OF SIGN LANGUAGE NAMES XVII NOTATION CONVENTIONS XIX 1. INTRODUCTION 1 1.1. LANGUAGE AND MODALITY 2 1.1.1. Simultaneity in sign languages 4 1.1.2. Iconicity in sign languages 10 1.2. THE USE OF SPACE IN SIGN LANGUAGES 17 1.3. THE STUDY OF VERBAL AGREEMENT IN SIGN LANGUAGES 23 1.4. LENGUA DE SIGNOS ESPAÑOLA (LSE) 25 1.4.1. LSE: historical background 27 1.4.2. LSE: sociolinguistic setting 29 1.4.3. Previous research on LSE 31 1.4.4. The LSE in this study 32 1.5. THE GOALS OF THIS THESIS 33 1.6. THE STRUCTURE OF THIS THESIS 34 2. THEORIES OF AGREEMENT 37 2.1. TWO APPROACHES TO AGREEMENT 38 2.2. TYPOLOGICAL APPROACH 42 2.2.1. Terminology 42 2.2.2. Controllers 43 2.2.3. Targets 45 2.2.4. Domains 56 2.2.5. Features and values 62 2.2.6. Conditions 72 2.2.7. Canonicity 74 2.2.8. Summary 77 2.3. THE MINIMALIST PROGRAM 78 2.3.1. Generativism: issues and developments 79 2.3.2. The architecture of the language faculty 85 2.3.3. Agreement and Agree 88 2.4. SUMMARY 94 3. AGREEMENT IN SIGN LANGUAGES 97 3.1. PRONOMINAL REFERENCE 98 3.1.1. Location assignment 99 3.1.2. Role shift 101 3.1.3. R-locus and space 103 3.2. AGREEING VERBS 107 3.2.1. Prototypical agreeing verbs 109 3.2.2. Backwards agreeing verbs 119 3.2.3. Single argument agreement 127 3.2.4. Summary 129 3.3. AGREEMENT AUXILIARIES 130 3.3.1. AUX 131 3.3.2. Auxiliaries derived from lexical verbs 134 3.3.3. PAM 137 3.3.4. Issue arising: what agreement auxiliaries tell us about agreement 140 3.4. NON-MANUAL AGREEMENT 142 3.4.1. Head tilt and eye gaze as markers of subject and object agreement 143 3.4.2. Non-manual agreement in role shift 147 3.4.3. Summary 149 3.5. DP-INTERNAL AGREEMENT 150 3.6. SUMMARY 153 4. METHODOLOGY 157 4.1. METHODOLOGICAL CHALLENGES: THE ELUSIVE NATIVE SIGNER 157 4.2. INFORMANTS 160 4.3. DATA COLLECTION AND MATERIALS 163 4.4. TRANSCRIPTION 165 4.5. DATA ANALYSIS 167 4.6. SUMMARY 167 5. AGREEMENT PHENOMENA IN LSE 169 5.1. PRONOMINAL REFERENCE 170 5.1.1. Location assignment in LSE 170 5.1.2. Role shift in LSE 175 5.2. AGREEING VERBS 179 5.2.1. Prototypical agreeing verbs 180 5.2.2. Backward agreeing verbs 183 5.2.3. Single argument agreement 184 5.3. AGREEMENT AUXILIARIES 189 5.3.1. AUX 190 5.3.2. Auxiliaries derived from lexical verbs: GIVE-AUX and BEAT-AUX 194 5.3.3. PERS 197 5.3.4. Summary 200 5.4. CONSTRAINTS ON VERBAL AGREEMENT 200 5.4.1. Semantic constraints on agreeing verbs 201 5.4.2. Phonological constraints on agreeing verbs 203 5.5. NON-MANUAL AGREEMENT 210 5.6. DP-INTERNAL AGREEMENT 213 5.7. DISCUSSION AND CONCLUSIONS 221 6. LSE AGREEMENT FROM A CROSS-MODAL TYPOLOGICAL PERSPECTIVE 227 6.1. CONTROLLERS 228 6.2. TARGETS 229 6.2.1. Verbs and auxiliaries 229 6.2.2. Other targets of agreement 231 6.2.3. Means of exponence 231 6.2.4. Multiple exponence 238 6.2.5. Summary 238 6.3. DOMAINS 239 6.3.1. Clause-internal agreement 239 6.3.2. Agreement beyond the clause 241 6.4. FEATURES AND VALUES 242 6.4.1. Gender 243 6.4.2. Number 243 6.4.3. Person 247 6.4.4. Other features: respect and case 253 6.4.5. Summary 255 6.5. CONDITIONS 256 6.6. CANONICITY 259 6.6.1. Applying Corbett’s criteria to spatial agreement in LSE 261 6.6.2. Applying Corbett’s general principles to spatial agreement in LSE 267 6.6.3. Other evaluations of the canonicity of sign language agreement 270 6.6.4. Summary 271 6.7. DISCUSSION AND CONCLUSIONS 272 7. FORMAL ANALYSES OF AGREEMENT IN LSE 277 7.1. LOCATION, IDENTITY AND LOCATING IDENTITY 278 7.1.1. The location of ϕ-features 280 7.1.2. The location of the identity feature 282 7.1.3. Optionality of the use of space 286 7.2. ACCOUNTING FOR SPATIAL AGREEMENT IN LSE 287 7.2.1. Location assignment 287 7.2.2. Verbal agreement 289 7.2.3. Pragmatic agreement 294 7.2.4. Summary 298 7.3. “DEFECTIVE” AGREEING VERBS IN LSE: AN OT ACCOUNT 300 7.3.1. “Defective” agreeing verbs in LSE 301 7.3.2. OT constraints 302 7.3.3. Applying the constraints 303 7.3.4. Extending the analysis to ISL data 308 7.4. ISSUES ARISING 309 7.4.1. Optionality (revisited) 309 7.4.2. Locative versus locus 310 7.4.3. Linearity 312 7.5. CONCLUSIONS 314 8. CLOSING REMARKS 317 8.1. WHAT THIS STUDY TELLS US… 318 8.1.1. …about LSE 318 8.1.2. …about sign languages in general 320 8.1.3. …about language 325 8.2. WHAT THIS STUDY DOES NOT TELL US 327 8.3. FUTURE DIRECTIONS 329 REFERENCES 333 LANGUAGE AND MODALITY: EFFECTS OF THE USE OF SPACE IN THE AGREEMENT SYSTEM OF LENGUA DE SIGNOS ESPAÑOLA (SPANISH SIGN LANGUAGE) (SUMMARY) 371 TAAL EN MODALITEIT: EFFECTEN VAN HET GEBRUIK VAN RUIMTE BINNEN HET CONGRUENTIESYSTEEM VAN LENGUA DE SIGNOS ESPAÑOLA (SPAANSE GEBARENTAAL) (SAMENVATTING) 377 LENGUAJE Y MODALIDAD: EFECTOS DEL USO DEL ESPACIO EN EL SISTEMA DE CONCORDANCIA DE LA LENGUA DE SIGNOS ESPAÑOLA (RESUMEN) 383 HIZKUNTZA ETA MODALITATEA: LENGUA DE SIGNOS ESPAÑOLA (ZEINU HIZKUNTZA ESPAINIARRA)-REN KONMUZTADURA SISTEMAK ESPAZIOAREN ERABILERAN DITUEN EFEKTUAK (LABURPENA) 389 xvii Abbreviations of sign language names ABSL Al-Sayyid Bedouin Sign Language (Negev Desert, Israel) ASL American Sign Language Auslan Australian Sign Language BSL British Sign Language CSL Chinese Sign Language DGS Deutsche Gebärdensprache (German Sign Language) DSGS Deutschschweizerische Gebärdensprache (Swiss German Sign Language) DTS Dansk Tegnsprok (Danish Sign Language) ESL Estonian Sign Language GSL Greek Sign Language ISL Israeli Sign Language IPSL Indo-Pakistani Sign Language KK Kata Kolok (Bali, Indonesia) KSL Korean Sign Language LIU Lughat al-Ishāra al-Urdunia (Jordanian Sign Language) Libras Lingua Brasilera de Sinais (Brazilian Sign Language) LIS Lingua italiana dei segni (Italian Sign Language) LSA Lengua de señas argentina (Argentine Sign Language) LSC Llengua de signes catalana (Catalan Sign Language) LSCol Lengua de señas colombiana (Colombian Sign Language) LSF Langue de signes française (French Sign Language) LSFB Langue de signes française de Belgique (Belgian French Sign Language) LSE Lengua de signos española (Spanish Sign Language) LSM Lengua de señas mexicana (Mexican Sign Language) LSQ Langue des signes québécoise (Quebec Sign Language) NGT Nederlandse Gebarentaal (Sign Language of the Netherlands) NS Nihon Shuwa (Japanese Sign Language) NZSL New Zealand Sign Language RSL Russian Sign Language SSL Swedish Sign Language TİD Türk İşaret Dili (Turkish Sign Language) TSL Taiwan Sign Language VGT Vlaamse Gebarentaal (Flemish Sign Language) xix Notation conventions The examples in different signed and spoken languages follow the (February 2008 version of the) Leipzig Glossing Rules (LGR), developed jointly by the Department of Linguistics of the Max Planck Institute for Evolutionary Anthropology and by the Department of Linguistics of the University of Leipzig. 1 Where examples are cited from other works, the transcription has been adapted to conform to the LGR as closely as possible. For ease of reference, the list of common abbreviations specified in the LGR is reproduced at the end of this section. Any abbreviations not included in the common LGR list are explained below the example in which they appear, and have been added to the list included here. The sign language examples include illustrative stills whenever possible and are transcribed using glosses in SMALL CAPS. While I have tried to maintain the conventions and abbreviations of the LGR, certain established glossing practices in the sign language literature have been maintained as standard, and are explained below:  Hyphens are used when more than one word is required to gloss a single sign: LOOK-AFTER Note that this differs from the LGR usage, for which a hyphen separates distinct morphemes.  Spatial modification of a sign is marked with a subscript. The subscript may indicate a location in the signing space (denoted by x, y, z or neut for the neutral location at the unmarked centre of the signing space) or on the signer’s body (denoted by 1): GROUPx Pointing or index signs (glossed as IX) invariably include a subscript to indicate the direction of the pointing. Subscripts are also used for referential indices, marking coreferentiality, but are distinguished from spatial modification by the index used: i, j and k (as opposed to x, y or z for locations in the signing space). For spatial modification involving movement between two points, as is the case for agreeing verbs, a subscript at the beginning of the gloss 1 Available on-line: www.eva.mpg.de/lingua/resources/glossing-rules.php xx denotes the initial location and a subscript at the end of the gloss denotes the final location: 1LOOK-AFTERx Note that this differs from LGR usage, which would use the “>” symbol for an affix that simultaneously expresses two arguments of a verb.  Fingerspelling is shown by individual, lowercase letters joined by hyphens: o-a-k  Classifier constructions are indicated by CL followed by a description of the form/meaning in parentheses: CL(group)y (Note that classifier constructions are invariably located in the signing space, so a subscript is included to show this.)  Relevant non-manual features are shown above the glosses of the signs, with horizontal lines indicating the scope or duration of the non-manual feature in question. The abbreviations used to categorize the non-manual features are in lower case and may describe function (e.g. “q” for a question marker) or form (e.g. “eyebrow raise”).  In most circumstances only a single gloss is given for the sign stream, but where the activity on each hand is relevant, the transcription includes a line for each hand, the upper line glossing the dominant hand and the lower line the non-dominant hand. When a given hand performs a hold (maintaining a given sign while the other hand continues to produce signs), a dashed line shows the duration of the hold: D hand ESTI BOYFRIEND COME IXmiddle-finger ND hand BUOYindex BOYFRIEND BUOYindex+middle----------  List buoys are shown by means of the gloss BUOY and include a subscript to indicate which fingers (of the non-dominant hand) are extended. When the dominant hand points at a buoy, the IX gloss is used with a subscript showing which finger (on the non-dominant hand) is being pointed at. (See examples above.) For examples with multiple signs, the relevant items are highlighted by bold face for the glosses, and a shaded background for the relevant images. Examples taken from recordings made for this study include the name of the recording followed by the time point at which the example occurs. xxi Common abbreviations of the Leipzig Glossing Rules This list includes the common abbreviations published in the February 2008 version of the Leipzig Glossing Rules that are used in this thesis, plus any further abbreviations that were required for the examples included (distinguished in boldface). 1 first person 2 second person 3 third person A agent-like argument of canonical transitive verb ABL ablative ABS absolutive ACC accusative ADJ adjective AGR agreement AOR aorist ART article ASP aspect AUX auxiliary BEN benefactive CL classifier COMP complementizer COMPL completive D hand dominant signing hand DAT dative DEF definite DEM demonstrative DEP dependent DIR directional DISTR distributive DTS direct theme sign DU dual ERG ergative F feminine FUT future G gender class GEN genitive HON honorific IIND independent indicative IMPF imperfect INAN inanimate INF infinitive INS instrumental INV inverse IX index LOC locative M masculine N neuter Nnon- (e.g. NSG nonsingular, NPST nonpast) NEG negation, negative NEUT neutral location in signing space NOM nominative OBV obviative P patient-like argument of canonical transitive verb PL plural PM phrase marker POSS possessive PRF perfect PRS present PRO pronoun/pronominal PROG progressive PRV preverbal PST past PU palms up Q question particle/marker REL relative RES resultative S single argument of canonical intransitive verb SBJ subject SBJV subjunctive SG singular TOP topic TR transitive 1 1. Introduction This thesis examines the nature of the agreement system in lengua de signos española (LSE – Spanish Sign Language). Within sign language linguistics, verbal agreement strategies are one of the most studied aspects of these languages. There are several reasons for such intense interest in the topic, and these have also motivated this study in the context of a specific sign language. Firstly, verbal agreement in sign language displays several unusual characteristics, such as restrictions on the number of verbs that show agreement, a typologically uncommon state of affairs. Secondly, verbal agreement in sign language makes use of strategies that are anchored to the visual-gestural nature of sign languages, and thus unavailable to spoken languages. Furthermore, the basic agreement mechanism is very similar (if not identical) across many unrelated sign languages, bringing into question the influence that the language modality may exert on a language’s structure and organization. The interaction of modality and language is the overarching theme for this thesis, and the research has been guided by far-reaching questions about the role sign language data can play in redefining our understanding of human language in general. These guiding principles are introduced in section 1.1 of this chapter. Section 1.2 concentrates on an aspect of sign languages that is a strong candidate for turning up modality effects since it is a mechanism that is unavailable to spoken languages: the use of space. Sign languages use space in different ways, and these are briefly described before limiting the discussion to one particular spatial device in section 1.3, namely verbal agreement. Section 1.4 gives a basic introduction to the specific language under investigation, LSE, and its most relevant characteristics, such as general sociolinguistic information and its relation to other sign languages. This section also includes an overview of previous research into the language. Section 1.5 articulates the specific research questions that provided the starting point for this study, and section 1.6 concludes this chapter by giving an outline of the remaining chapters. 2 Introduction 1.1. Language and modality This study focuses on a specific aspect of a specific sign language, but is couched in a much broader perspective. Firstly, the LSE data will be compared with data from other sign languages and also spoken languages to provide a typological context for agreement in LSE. Furthermore, the wider consequences of the findings for linguistic theory in general will be examined by taking a step back to see the bigger picture. Sign languages offer the unique opportunity to look at the effect of modality on language: Why should we be interested in whether specific aspects of linguistic structure might be attributable to the particular properties of the transmission channel? Exploration of modality differences holds out the hope that we may achieve a kind of explanation that is rare in linguistics. Specifically, we may be able to explore hypotheses that this or that property of signed or spoken language is attributable to the particular constraints that affect that modality. (Meier 2002: 5) If linguistic research limits itself to spoken languages and the proposals for the fundamental nature of language are based solely on spoken language data, it will be impossible to know whether recurrent properties reflect general design characteristics of human languages or are merely due to the vocal/auditory medium. By including sign languages in the linguistic program, the variable of modality is introduced and we may hope to distinguish core language properties from modality effects. Can we hope to find modality effects by comparing signed and spoken languages, or are they essentially the same? Although the field has been marked by a tendency to highlight the similarities between spoken and signed languages – due in large part to a need to socially dignify sign languages and to justify their inclusion within the discipline of linguistics (Woll 2003) – recent research has started to look for possible differences between signed and spoken languages (Meier, Cormier & Quinto-Pozos 2002; Vermeerbergen 2006). This “sign differential” view is an attempt to study sign languages in their own terms without applying inappropriate concepts or imposing models developed in the context of spoken languages. This approach to sign language research is closely linked to the idea that sign languages are qualitatively different to spoken languages and have different organization and structure. These dissimilarities are due to the distinct modalities of signed and spoken languages and the specific sociolinguistic context of sign languages, especially their relative youth (Aronoff, Meir & Sandler 2005). The modality differences may be due to the contrasting nature of the articulators used for language Language and modality 3 production and the perceptual systems used for language comprehension, and the resulting potential for iconicity that arises from the use of space (Meier 2002, 2012). The articulators employed by sign languages give rise to possibilities of simultaneity (section 1.1.1) and, together with the use of space, to a greater exploitation of iconicity (section 1.1.2); the use of space itself, the main focus of this thesis, will be looked at in section 1.2. The notion that modality shapes language coupled with the observation that many different (and unrelated) sign languages have similar structures and make use of analogous mechanisms (such as classifier constructions and non-concatenative morphology) leads to the suggestion that sign languages are not only different to spoken languages because of their different modality, but also that they are similar to one another because of their shared modality. However, it is important to bear in mind that sign languages show greater variation between themselves than was once realised and growing research on a wider range of sign languages confirms that (superficial) universal properties are hard to come across (Perniss, Pfau & Steinbach 2007). Furthermore, work on non-western sign languages and particularly “shared sign languages”, which exist in sociolinguistic contexts quite unlike that of most western sign languages studied to date, have revealed greater variability across languages in the visual gestural modality. (For an overview see Nyst 2012 and de Vos & Pfau 2015.) These differences between sign languages may be accounted for in terms of diverse factors such as modality, typology and parametric variation within the framework of Universal Grammar, in the same way that linguistic diversity is explained for spoken languages (Hohenberger 2007). This brings us to the alternative to the “sign differential” view: the “sign same” position holds that sign languages are essentially the same as spoken languages, both being expressions of the underlying language faculty that has fixed core properties (Pinker & Jackendoff 1995). Of significance is the nuance of underlying similarity: signed and spoken languages have undeniable differences, but there are enough similarities to claim that a common computational component serves both (Lillo-Martin 2001, 2002, 2006). The concept of a specific language component is closely associated with generativist linguistics and the Chomskian tradition which claims that language is an innate human faculty that exists as a specific cognitive module in the brain (Chomsky 1965; Fodor 1983; see section 2.3 of the next chapter for a brief overview of generativist linguistics). We now turn to specific aspects of modality differences between signed and spoken languages (simultaneity and iconicity) before moving on to the issue of space in sign languages. 4 Introduction 1.1.1. Simultaneity in sign languages The articulators used to produce sign language are radically different to those employed by spoken language. The most salient difference is a question of scale: the hands, arms, upper body and head are much larger and occupy a greater volume than the vocal apparatus. This gives them much greater visibility and allows them to make use of space in a way that the larynx, epiglottis, tongue and lips do not, as we shall see in section 1.2. Furthermore, together with this macro-scale visibility, the articulators are relatively independent of one another and can perform different movements at the same time. This opens up the possibility for simultaneity in sign languages. Sign language production may be divided into two main channels: the manual and the non-manual. The manual component is articulated by the hands, and the non-manual component is expressed by the head (tilts, nods, shakes), the eyebrows (raised, frowning), the eyes (gaze direction, blinks, aperture), the nose (wrinkling), the mouth (mouthing), the shoulders (raised) and upper body (tilts, turns). I will look first at simultaneity within the manual component, and then at the non-manual component. The fact that sign language is articulated by the hands, of which there are two, makes it possible to be doing two different (linguistic) things at the same time. However, it is not true to say that signers have two independent articulators equivalent to having multiple voices, like those of Willie from the cartoon The whale who wanted to sing at the Met, who could sing duos with himself. On the whole, for most signing production, the hands work in coordination: one (non-dominant) hand is subjugated to the other (dominant) hand, and there are restrictions on what the non-dominant hand can be doing according to the activity of the dominant hand (Battison 1978). This holds true for most lexical signs (those with a fixed form whose meaning is not entirely subject to context), but under certain circumstances the hands may act with a greater degree of autonomy. A taxonomy of different types of simultaneous construction is proposed by Miller (1994) and here we shall look at two broad groups of simultaneous bimanual constructions: co-occurring lexical signs and classifier constructions. The first type of simultaneity occurs with one-handed signs. Some signs do not make use of the non-dominant hand, and in the case of some twohanded signs the non-dominant hand may be suppressed (Battison 1974; Padden & Perlmutter 1987; Brentari 1998). With this type of sign, it is possible for each hand to articulate a different sign at the same time, which may be compared to uttering two words simultaneously, as can been seen in the British Sign Language (BSL) examples in (1). Language and modality 11 sense) and its referent. An example is the sun symbol used on a weather forecast map; both sign and referent are round and bright yellow. Diagrammatic iconicity is a correspondence between parts of a representation and parts of the thing it represents. An example is the number of lines on a symbol to represent the strength of the wind; the relation between the wind barbs corresponds to the relation between different types of wind, such that the more bars there are, the stronger the wind (as exemplified in figure 1.1). In the linguistic realm, diagrammatic iconicity is present in the correspondence between the temporal order of orations and that of the events described, captured by Jakobson’s (1965: 26) classic example ‘veni, vidi, vici’. 6 These definitions of iconicity are based on Pierce’s (1932) seminal work on semiotics. 7 round bright yellow more bars stronger wind PERCEPTUAL FEATURES OF SYMBOL RESEMBLE PERCEPTUAL FEATURES OF REFERENT RELATION BETWEEN PARTS OF SYMBOL ANALOGOUS RELATION BETWEEN PARTS OF REFERENT Figure 1.1. The relationship between form and meaning in imagistic (left) and diagrammatic (right) iconicity. Iconicity is closely related to the notion of motivation, and this is perhaps one of the reasons why sign languages were excluded from linguistic study for so long. 8 The Saussurean dogma of the arbitrariness of the linguistic sign as a defining feature of language led to signed languages being dismissed as mere pantomime. And indeed, the forms of signs do tend to show greater motivation than those of words. Considering that much of what language is used to talk about refers to the visual world, it is not surprising that a visual language shows a great deal of motivation in the form of its signs. Conversely, if we focus on the domain of sound related concepts, spoken languages show a much greater degree of motivation by means of onomatopoeia and sound symbolism (Hinton, Nichols & Ohala 1994). 6 For the development of the concept of diagrammatic iconicity in spoken languages, see the papers in Haiman (1985). 7 Peirce also included a third type of iconicity: metaphors. 8 For an overview of the history of sign linguistics, see McBurney (2001). 12 Introduction Equally, we seem to have a natural bias for considering iconicity in visual terms and tend not to pick up on correspondences between form and meaning in other dimensions, such as temporal organization (in this respect, for sign languages see Wilbur 2008). Much of the literature on iconicity in sign languages deals with imagistic iconicity at the word/sign level (e.g. Klima & Bellugi 1979; Pizzutto & Volterra 2000; Pietrandrea 2002; Wilcox 2004; Ormel, Hermans, Knoor & Verhoeven 2009; Perniss, Thompson & Vigliocco 2010; Baus, Carreiras & Emmorey 2013). 9 The fact that the sign for a given concept may bear a visual connection to its referent is unquestionable, as can be seen from the LSE examples in (8). The examples in (8a) and (b) are transparent enough that they would appear in pantomimic gestures for the same meanings; in (8c) BILBAO depicts the traditional musical instruments (the txistu, a one-handed flute, and the danbolin, a drum) typical to the region where the city is located LSE (TZ) (8) a. b. c. CAR COLD BILBAO However, there are two important observations to be made. Firstly, although signs may show a certain degree of visual motivation, this does not exclude some level of abstraction. In (8a), the sign CAR uses metonymy to associate (the action of handling) a part of the referent with the whole; (8b) COLD associates an action typically used to counteract the effects of a physical sensation with the concept, in what may be described as a type of indirect synaesthesia; (8c) BILBAO depicts cultural artefacts associated with the referent. The process for the creation of an iconic sign may be broken down into various stages: conceptualizing, image selection, schematizing and encoding (Taub 2001). This relates to the second point concerning iconic signs: even though a given sign may have a (high) degree of visual motivation, this does 9 Work which looks at the role of (imagistic) iconicity at the discourse level in sign languages includes Sallandre & Cuxac (2002) and Russo (2004). Language and modality 13 not make its meaning transparent and does not rob it of all arbitrariness. The fact that sign-naïve subjects fail to guess the meaning of signs above chance confirms the relative opacity of the forms (Klima & Bellugi 1979), and (8c) demonstrates that specific cultural knowledge may be required to decipher the visual motivation behind the form of a sign. Furthermore, the variation in lexical form across sign languages, as illustrated in (9) by the different signs used by three different sign languages – LSE, Australian Sign Language (Auslan) an Colombian Sign Language (LSCol) – for the same meaning, confirms that signs do have an element of the arbitrary. More recently, sign researchers have begun to examine the role of diagrammatic iconicity. Recall that in the case of diagrammatic iconicity there is no need for the sign to resemble the referent, but rather the parts of the sign reflect a relationship between the parts of the referent (see figure 1.1). For spatial descriptions, sign languages may make use of space to describe location and motion. The placement and movement of the signs relative to each other corresponds to the location and motion of the referents to each other; the signing space acts as an map and is exploited topographically (Emmorey, Corina & Bellugi 1995). (9) The sign CAR in three different sign languages. (Image in (b) reproduced with kind permission from the author.) a. b. c. LSE (TZ) Auslan (Victoria dialect) (Johnson 1998: 285) LSCol (INSOR 2006: 310) This topographic use of space normally occurs in conjunction with classifier constructions – see (4) for such a structure in DGS – but may also occur with lexical signs, as demonstrated by the LSE description of the water cycle in (10), which mixes classifiers with lexical signs such as SUN and RAIN within a diagrammatic spatial framework. The direct isomorphic mapping that topographic space establishes between the spatial relations of the signs and 14 Introduction those of the referents creates a perceptual similarity that is reminiscent of imagistic iconicity. However, note that the space is not the sign itself but forms part of the relationship between the signs, and that it is this (spatial) relationship that is analogous to a (spatial) relationship between the referents. LSE (TZ) (10) (Use of space throughout the discourse.) D hand CL(liquid) SUN CL(evaporate) CLOUD RAIN ND hand CL(liquid)------hold---------- CL(evaporate) CLOUD RAIN D hand CL(solid) SUN CL(flow) CL(liquid) ND hand CL(solid)------hold--------- CL(surface)—hold----- ‘The water is evaporated by the sun and rises into the air as vapour, where it forms clouds. These clouds then move over the landmass and become denser, eventually leading to precipitation. The water freezes on the mountains but is then melted by the sun and flows down the mountain back into the sea.’ Apart from the topographic use of space, a further instance of diagrammatic iconicity in sign languages is identified by Wilbur (2008), whose Event Visibility Hypothesis states that the path movement of a predicate sign maps onto the duration of the event being described. In this case, the correspondence is between spatial relationships and temporal relationships. Such instances of diagrammatic iconicity highlight the motivated nature of Language and modality 15 certain mappings employed by the language system and recent work suggests that this motivation may be driven by universal perceptual mechanisms (Strickland, Geraci, Chemla, Schlenker, Kelepir & Pfau 2015). Nevertheless, this is not the end of the story: although iconicity may be present in a language, its role may be to feed possible forms into the system, which then conventionalizes and grammaticalizes these forms so that they fit into the linguistic system (Wilbur 2008). Work on the grammaticalization paths of sign languages suggests that some linguistic forms may derive from iconically motivated gestures (Pfau & Steinbach 2006a). That sign language forms are abstract, conventionalized symbols is confirmed by evidence from acquisition studies of sign languages: imagistic iconicity does not affect the acquisition of signs in the early stages of language development (Orlansky & Bonvillian 1984) and the acquisition of the verbal agreement system is guided by the morphology rather than the iconicity of the forms (Meier 1987). Furthermore, the classifier system, with its diagrammatically iconic use of topographic space is rule-governed and operates on systematic linguistic principles (Supalla 1982, 1986; Zwitserlood 2003). Iconicity is present but this does not necessarily alter the workings of the linguistic system. When considering the role of iconicity in spoken languages, we have already seen that phenomena such as onomatopoeia are instances of imagistic iconicity; as far as diagrammatic iconicity is concerned, there is a growing body of work that teases out the diagrammatic relations in language structure at the levels of morphology (Bybee 1985), syntax (Haiman 1985) and discourse structure (Karrabæk 2003). Just as the notion of iconicity rests on the intuition that the structure of language reflects the structure of experience (Croft 2003: 102), it could be argued that the generativist stance that syntactic structure maps onto formal semantic structure resembles an iconic relation in its isomorphism. However, the status of iconicity in language structure is disputed, and it has been suggested that principles based on iconic considerations may be due to other factors such as frequency of use, or that the concept of iconicity is better expressed in terms of notions of economy and distinctiveness (Haspelmath 2008). This calls to mind Wilbur’s suggestion that iconicity may make motivated forms available, but the language system then grammaticalizes these forms. Before closing this section on iconicity, I wish to return briefly to the notion of arbitrariness. Aside from the debate over the role of iconicity in language structure, the need for an arbitrary relation between form and referent has lost its foothold. Firstly, we have seen that sign languages show a relatively high degree of motivation in the forms of signs. Furthermore, studies in 16 Introduction synaesthesia show that the naming of objects is not a completely arbitrary matter: the so-called bouba/kiki effect (see figure 1.2) demonstrates that there is some sort of underlying correspondence between the visual appearance of the referent and the form of the linguistic sign even in spoken languages (Ramachandran & Hubbard 2001). The authors claim that this effect may go towards explaining the occurrence of sound symbolism, but in the context of sign languages the effect may account for the high degree of motivation. Spoken languages have a long history and have undergone thousands of years of evolution; the changes in the linguistic system have led to an arbitrary relationship between sign and referent. Conversely, (as we shall see below in section 1.4.2) sign languages are relatively young languages and their evolution is stunted by the particular sociolinguistic circumstances in which they exist. As a result, the naming processes have occurred much more recently 10 and so the motivational link between referent and form is still present. As such, arbitrariness may be a property of old languages, but not of younger languages; the factor common to both types of language is the fact that linguistic sign is symbolic (Sandler & Lillo-Martin 2006: 499). Figure 1.2. The bouba/kiki effect. When asked to choose between two names for these two different shapes, 95% of subjects choose “bouba” for the rounder shape and “kiki” for the jagged shape. (Adapted from Ramachandran & Hubbard 2001: 19.) In this section we have looked at the role of iconicity in sign languages. Although sign languages show a high degree of both imagistic and diagrammatic iconicity, both types of motivated form-meaning relationship also exist in spoken languages. Furthermore, the non-arbitrary character of a form does not necessarily detract from its linguistic status, and the important question is whether the forms are subject to the rules of the language system. 10 Naming processes may even reoccur with each generation. See section 1.4.2 of this chapter for the suggestion that sign languages undergo a constant process of creolization. The use of space in sign languages 17 As languages in the visual-gestural modality, sign languages are more susceptible to creating and using visually motivated forms (and this may be reinforced by the relative youth of the languages) as well as mechanisms such as topographic space and event visibility (Wilbur 2008). Notice that both of these iconically driven devices involve the use of space and that, once again, the assessment of modality effects has led us to considerations of the use of space in sign languages. The next section examines this topic in detail. 1.2. The use of space in sign languages Sign languages are expressed in the visual modality and, as we have seen, make use of articulators very different to those of spoken languages: the hands and upper body, including shoulders, head and face. Signs are either produced on the body or near the body in the signing space, which, in most sign languages, is approximately the quarter-spherical volume just in front of the signer (see figure 1.3). Figure 1.3. The signing space occupies approximately a quarter-spherical volume in front of the signer. This means that sign languages can make use of space in a way that spoken languages cannot, and this use of space is pervasive throughout the language: In speech, the acoustic signal derives from, but is different from, the motion of the articulators (visible information on the lips is extremely limited with respect to the whole phonological inventory). In signing, the visual signal – the hands moving – is the motion of the articulators, that is, what is seen is the temporal dynamics and spatial location of hand movement. The linguistic system depends on the visual perceptual system to process the necessary distinctions. (Wilbur 2008: 218) At the phonological level, space forms one of the basic building blocks used to make up an individual sign: current phonological models for sign language 18 Introduction include some sort of location feature that refers to the point or points in the signing space where the sign is produced (Sandler 1989; Brentari 1998; van der Kooij 2002). (11) shows an example of a minimal pair of signs in LSE that differ only in the location of each sign while the handshape, movement and orientation of both signs are identical: both BROWN and AMAZED involve flicking the middle finger off the thumb with the palm facing away from the body, but the former is articulated in neutral space while the latter is produced in front of the face (normally in front of the forehead). LSE (TZ) (11) a. b. BROWN (Basque dialect) AMAZED Also at the phonological level, many signs may include a (path) movement feature in their lexical specification: movement which, of course, occurs through space. 11 Examples of signs with a path movement are given in (12). The sign THEN describes a circle in the neutral space on the signer’s ipsilateral side; the sign BASQUE-POLICE involves a short arc movement of the hand while it is in contact with the contralateral side of the signer’s chest. A sign may also be moved through the signing space in order to add information to the sign, making the movement component morphemic in nature. This strategy is exemplified by classifier constructions, which employ a movement morpheme to express predicates of existence or motion (Supalla 1982; Emmorey 2003), as illustrated in the examples in (13): (13a) describes the relative positions of three chairs, and (13b) gives information about the movement of the cat relative to the signer. 11 Phonological movement may also be internal – in Sandler’s (1989: 92) terms – or local – in Brentari’s (1998: 130) terms – involving a change in the handshape or the orientation of the hand but not a path movement through space. Such movement may be seen in the signs in (11) above. The use of space in sign languages 19 LSE (TZ) (12) a. b. THEN BASQUE-POLICE LSE (TZ) (13) a. CHAIR THREE CL(chair)left CL(chair)middle CL(chair)right ‘There are three chairs: one on the left, one in the middle and one on the right.’ b. YESTERDAY CAT MOUSE CL(bite) CL(walk) CLx(walk)1 ‘Yesterday my cat brought me a mouse it had hunted.’ As we saw in section 1.1.2 when looking at the issue of iconicity in sign language, the signing space may be exploited topographically in order to provide spatial descriptions. (For example, the spatial positioning of the signs in (3), (10) and (13) is analogous to the spatial positioning of the referents.) The expression of spatial information in sign languages and the use of space that this involves provide insight into the mechanisms and constraints at work in the visual-gestural modality (see Perniss 2007 for DGS). However, the 20 Introduction focus of this thesis is the notion of agreement and the use of space in the expression of agreement relations; as such, spatial descriptions (and the associated use of topographic space) will be referred to only where relevant. LSE (TZ) (14) (Use of space throughout the discourse.) SIGNx SPANISHy IXy IXx xLINKy SAME NOT ‘LSE and Spanish are separate languages that are not the same as each other.’ Space may also be used at the discursive level, with different discourse topics or themes being associated with different areas of the signing space (Emmorey, Corina & Bellugi 1995). In (14), the signer contrasts two languages (Spanish and LSE) by associating each with either side of the signing space. Since neither of these signs is body-anchored, they are articulated directly at the different locations in the signing space. LSE Spanish Lengua de signos española (LSE) 27 This section offers a brief introduction to various aspects of the language. Section 1.4.1 provides information on the historical background of LSE, which is relatively well historically documented compared to other sign languages. The current sociolinguistic situation of LSE, essential to understanding many of the factors that condition the language, is described in section 1.4.2. An overview of previous linguistic work on LSE is provided in section 1.4.3, and the last section, 1.4.4, identifies the characteristics of the specific variant of LSE analysed in this thesis. 1.4.1. LSE: historical background LSE is used in the Deaf community throughout most of Spain, including the offshore territories of the Canary Islands, Ceuta and Melilla, with the exception of Catalonia, where LSC is used (see above). The origins of the language are not known, although some sort of sign language has been in use in Spain for at least four hundred and fifty years; Spanish Sign Language is in the privileged position of having a relatively rich body of historical literature dating from the mid-sixteenth century onwards (cf. Lasso 1550/1919). Many of these documents are freely available as digital facsimiles in the Biblioteca de Signos, which forms part of the Biblioteca Virtual Miguel de Cervantes hosted by the University of Alicante. 18 Needless to say, the information offered by these texts is somewhat erratic and subject to the authors’ prejudices and intent. Many of the writings, such as Bonet’s Reduction de las letras y arte para enseñar a ablar los mudos (1620), deal with the education of the deaf, and specifically how to “improve” the sign language by making it more similar to the spoken language. Furthermore, what little information these texts offer is in the form of written descriptions and any graphic representations are inevitably of the manual alphabet, which reveals next to nothing about the sign language itself (see figure 1.5). Even so, these historical texts provide unquestionable evidence that there has been a community of sign language users in Spain for several centuries. Another matter is how similar the sign language of the sixteenth or seventeenth century was to present day LSE. The sociolinguistic characteristics of sign languages give rise to a great deal of variation both historically and geographically, and this matter is dealt with in the next section. Along with the documented history of the sign language used by deaf people in Spain over the last few centuries, there is a rich oral tradition among the Deaf community with historical figures such as Pedro Ponce de León or Juan de Pablo Bonet featuring in story-telling. Although this practice is less 18 http://www.cervantesvirtual.com/seccion/signos/ 28 Introduction prevalent among the younger generations of signers, knowledge of the history of the Deaf Community is regarded as a means of forging social identity. Conversely, some authors have criticised the apparent mythification of certain characters and episodes (cf. Gascón Ricao & Storch de Gracia y Asensio 2004). Figure 1.5. The manual alphabet taken from a textbook for teaching deaf students (Ballesteros & Fernández Villabrille 1845). It has been claimed that LSE belongs to the sign language family of SouthWest Europe (Anderson 1979 cited in Woll, Sutton-Spence & Elton 2001: 26), which includes French Sign Language (LSF) and Sign Language of the Netherlands (NGT). However, as mentioned above, the origins of LSE are unclear. Apart from a current lack of sufficient knowledge on specific sign languages to establish genetic links between them, the discipline of historical linguistics has not been developed for sign languages, 19 thus making it 19 Exceptions are Frishberg (1975), Woll (1987), Wittmann (1991), Janzen & Shaffer (2002) and Wilcox (2004). For an overview see McBurney (2012). Lengua de signos española (LSE) 29 difficult to go beyond speculation about the relations between the older sign languages or even the existence of some “European Proto Sign Language”. 20 1.4.2. LSE: sociolinguistic setting As is the case for most western sign languages, the sociolinguistic setting of LSE is quite complex: the language exists in a permanent state of bior multilingualism, has no written form, has been subject to institutionalized oppression and is visual-gestural in form. Leaving aside the difference in modality, many of these characteristics are typical of minority spoken languages. What really sets sign languages apart is their lack of generational continuity: the vast majority of deaf children are born to hearing families, and as such they will not normally be able to acquire sign language in a natural setting. The figure often cited in the literature is that between 5-10% of deaf children are born into deaf families (Schein & Delk 1974), but recent work suggests that this holds only for the United States, where a particular strain of genetic deafness and a certain degree of endogamy within the deaf population has led to favourable conditions for multigenerational deaf families (Nance, Liu & Pandya 2000). The figure may be much lower for other countries (Johnston 2006), Spain included (Costello, Fernández & Landa 2008). This situation means that there are extremely few native signers in the signing community. Hence, from the point of view of linguistic research, the methodology has to be adapted to these peculiar conditions. The research methods and data collection techniques used in this study are described in chapter 4. The fact that so few deaf children are born into an environment that is conducive to their acquiring sign language means that many learn a sign language at a late age (either upon starting a formal education or even later, when they have left school) and this has a huge impact on the language itself. As might be expected under such circumstances, each sign language shows a fairly high degree of variation. Furthermore, it has been proposed that the generational discontinuity causes sign languages to undergo a continual process of creolization (Fischer 1978; Meier 1984; Aronoff, Padden, Meir & Sandler 2003). Whether or not the analysis of sign language as some sort of creole is correct (Lupton & Salmons 1996), the low level of native-like 20 This clearly is not the case for sign languages which have developed in the context of more recent historical processes such as colonialism, as is the case for the attested proximity between Auslan – Australian Sign Language – and BSL – British Sign Language (cf. Johnston 2003). 30 Introduction acquisition amongst the users certainly affects the language itself (Costello, Fernández Landaluce, Villameriel & Mosella 2012). A clear example of this is the role that educational policy can play in the development of the language: since many individuals learn sign language at school rather than at home, the type of language they learn is subject to the whims of educational policy (Bouvet 1990). In most western countries, the policy for Deaf education has been subject to radical changes in the last 150 years and this is no less true of Spain. Teaching practice and philosophy have gone from sign-based methods to a long period of oralism (during which the use of sign language was discouraged or actively punished), to the more recent reintroduction of sign language as part of a bilingual/bicultural pedagogy accompanied by mainstreaming (Plann 1997; Minguet Soto 2001). This tendency to insert deaf pupils in ordinary schools is significant because the residential deaf schools were traditionally the seat for sign language learning and often gave rise to variants that formed the main regional dialects of a national sign language (Kyle & Woll 1986). The disappearance of the residential schools has meant the loss of the foci of different dialects. The dialectal variation of LSE has been studied by means of lexical comparison using a Swadesh type word list in the glottochronology tradition (Gudshincksy 1956; Swadesh 1972) and mutual intelligibility tests (Parkhurst & Parkhurst 2007). The study looked at sign language use in 18 different parts of Spain and the findings confirmed that certain lexical differences do exist from one region to the next, but the level of mutual intelligibility between signers from different areas is well above the 75% threshold usually applied to distinguish different spoken languages (SIL 1991: 45, cited in Parkhurst & Parkhurst 2007: 46). Furthermore, the findings distinguish between LSE and LSC, each with their own internal dialectal variation. The Parkhurst study also provides a general overview of the situation of sign languages in Spain, including details on the role of Deaf Schools and Associations in the development of language varieties. For further information on the signing communities of Spain, see Vallverdú (2001) and Quer, Mazzoni & Sapountzaki (2010); for more specific sociolinguistic information, see Minguet Soto (2001). For the sociolinguistics of sign languages in general, see Lucas (2001). LSE was granted a certain degree of official recognition by a law passed in late 2007 that set out to regulate the rights and communication options of deaf individuals, though it remains to be seen what impact this law will have on sign language use and the development of the language in the long term. 21 21 The full Spanish text of the law is available on-line: Lengua de signos española (LSE) 31 1.4.3. Previous research on LSE Taking a broad view, the description of LSE appears to conform to many of the findings for other western urban sign languages that have been studied: the language has a classifier system and makes use of non-manual features, to cite just two of the features that typify sign languages. The relative similarity between sign languages leads to the danger of over-generalizing findings from one sign language to others, but scientific method demands that each object of study be examined in its own right. Although the large body of research on ASL (the most studied sign language to date) may inform work on lesser studied languages, it is important for those languages to be investigated in order to find out what their own peculiarities are. So while LSE may look fairly similar to other sign languages, there is still a need to carry out research, if only to confirm first impressions. This section presents an overview of the research work that has already been done on LSE. There are several dictionaries of LSE, the first published in the midnineteenth century (Fernández Villabrille 1851). Since 1957 a dictionary has been published under the auspices of the National Association of Deaf People, CNSE, (Marroquín Cabiedas 1957, cited in Vicente Rodríguez et al. 2008, and Pinedo Peydró 1981) and in recent years specialized thematic dictionaries have been created (Fundación CNSE 2002-2003). Motivated by the 2007 law that provides certain legal recognition for LSE and lays down provision for a standardization process for the language (see previous section), CNSE has published a normative dictionary with over 4,000 entries (Fundación CNSE 2008). 22 All these dictionaries are limited to single sign entries and provide no grammatical information about LSE. There has been very little modern linguistic analysis of LSE, and this thesis forms part of a growing body of work that documents the language. Previous work on the language includes five doctoral theses: Rodríguez González (1990) presents a general linguistic analysis of LSE with respect to the structure of Spanish; Fernández Soneira (2004) on quantification in LSE; Iglesias Lago (2006) on non-manual features to express modality; Gras Ferrer (2006) on the sociolinguistic status of sign languages in Spain; Gutiérrez Sigut (2008) on the role of phonological features in the processing of LSE. Currently there are also several doctoral dissertations on LSE in progress at various Spanish universities. Published work on LSE is limited to a learners’ grammar (Herrero Blanco 2009) and a collection of articles published by the National Association http://www.boe.es/boe/dias/2007/10/24/pdfs/A43251-43259.pdf 22 This dictionary can be accessed on-line: http://www.fundacioncnse.org/tesorolse 32 Introduction of Deaf People (CNSE 2000), a study of register in LSE (Chapa Baixauli 2001) and a general textbook dealing with a range of issues that draws on literature of other sign languages (Minguet Soto 2001). There are also several articles (almost exclusively in Spanish 23 ) published by the various research groups working on LSE: the main groups are to be found at the University of Vigo and the University of Alicante. The Alicante group has also published various materials on-line, including a bilingual (LSE-Spanish) basic grammar of LSE (Herrero et al. 2005). 24 1.4.4. The LSE in this study This study looks specifically at the variety of LSE used in the Basque Country, a region in the north of Spain (shaded solid in figure 1.4). As noted above, LSE shows a certain amount of variation but this does not affect intelligibility between users from different areas. In particular, signers from the Basque Country can easily understand and be understood by signers from other parts of Spain. Furthermore, the members of the signing community of the Basque Country consider their language to be LSE. Bilbao, the largest city in the Basque Country, was home to a large residential deaf school (which in the last 20 years had been reduced by mainstreaming to a unit for pupils with mixed special needs within an ordinary school and recently closed down altogether). The LSE variant of the Basque Country is given some uniformity by the fact that many older signers learnt to sign there. Further details of the data collected for this study and the signers who provided the data will be provided in chapter 4. It should be emphasized that there is no Basque Sign Language, though the Basque Country does have two main spoken languages: euskera, or Basque, a language isolate of unknown origin, and castellano, or Spanish, a Romance language used throughout all of Spain. This spoken language bilingualism adds to the complex sociolinguistic background of the LSE variant used in the Basque Country, but this issue will barely be touched upon in this work. Most Deaf people in the Basque Country have Spanish as their main spoken language (be it their first or second language), but the promotion of Basque in the compulsory education system has meant that the dominant spoken language for some younger signers is Basque (normally for those whose families are Basque speakers). Given the disparity between 23 Articles on LSE published in English include Cabeza Pereiro & Fernández Soneira (2004) and Herrero Blanco & Salazar García (2005) in addition to the dialect study mentioned in section 1.4.2: Parkhurst & Parkhurst (2007). 24 Unfortunately the format of the on-line grammar is outdated and it has been difficult to access the contents since 2009. The goals of this thesis 33 Spanish and Basque, this situation presents a fascinating opportunity to examine language contact phenomena between a sign language and two very different spoken languages, but I leave this area to future research. 1.5. The goals of this thesis Having outlined the general topic and introduced the individual language that this thesis deals with, I now turn to the specific objectives of this research work. What mechanisms does LSE use for agreement, and to what extent are they the same as or different to agreement mechanisms employed by other sign languages? The first task in hand is largely descriptive as I set out to describe the verbal agreement process in LSE, looking at its phonological manifestation and focusing on any peculiarities it may have when compared with what has already been described for other sign languages. Agreement for sign languages, based on work on a variety of sign languages, is described in chapter 3; the phenomenon in LSE is described in chapter 5. The relevant comparisons are made throughout chapter 5 and in the concluding section of that chapter. Even though the details need to be provided in the description of the LSE data on agreement, we know that this process involves the use of space and the association between a locus and a referent (see sections 1.2 and 1.3 above). In the sign language literature, (certain cases of) this spatial mechanism are referred to as “agreement” and considered to be analogous to the same phenomenon in spoken language. One of the aims of this thesis is to assess how valid this identification is. Are the spatial mechanisms employed by LSE comparable to the agreement mechanisms in spoken languages? This will be done by using two different linguistic traditions as yardsticks for the sign language data. Firstly, a typological approach provides a broad view that defines agreement in terms of the different structural elements that play a role in the process, and, furthermore, offers a rich overview of the different options that exist in typologically diverse languages of the world. Secondly, minimalist syntax, from the generativist tradition, offers a technical and highly specified notion of agreement in terms of structural relations and syntactic operations. These different – but complementary – approaches provide the means to hold up the LSE data against data from other spoken languages, and theories developed based on spoken language data. 34 Introduction Evaluating the spatial agreement process in this way will also offer the opportunity to develop a formal characterization of the phenomenon. In describing and analysing the data, it will become evident whether or not spatial agreement in LSE fits into existing models, and what, if any, adjustments are necessary to accommodate the model to the data. Can spatial agreement in LSE be given a formal characterization? These, then, are the three research questions that drive the research laid out in this thesis. They can be summarized as an attempt to compare LSE agreement to what is known about other sign languages and about spoken languages to decide whether the phenomenon can correctly be characterized as agreement using (and, if necessary, adapting) current models. 1.6. The structure of this thesis This thesis is structured in the following manner. I begin by providing the theoretical background for agreement, from the point of view of general (spoken language) linguistics and sign languages, respectively. After describing the methodology employed, I focus on agreement in LSE based on the data collected for this study and provide a detailed account of how agreement is manifest in this language. I then apply two different approaches (typological and generativist) to analyse the LSE data with a view to evaluating how well agreement in LSE fits into the cross-linguistic landscape. I also provide a formal account of agreement in LSE based on the idea of a basic spatial agreement mechanism. I conclude by taking a step back to consider what this spatial agreement mechanism in LSE can tell us about modality effects and language in general. Chapter 2 presents an overview of agreement as a linguistic phenomenon from two different frameworks: linguistic typology, which sets out to describe data from as broad a range of languages as possible; and Generative Grammar, which offers a set of concepts and mechanisms that provide a detailed syntactic account of the workings of agreement. Although these two frameworks represent quite different approaches, I justify using both as complementary methods, each of which contributes its own benefits. Chapter 3 provides an overview of the phenomenon of agreement as it has been described in the sign language literature. This involves looking first at how locations in space are used for reference, including the process of location assignment. The overview takes in phenomena that have previously been analysed as agreement (namely, agreeing verbs, backwards verbs, agreement auxiliaries, and non-manual agreement). Additionally, I also describe and consider similar uses of space as possible candidates of a spatial The structure of this thesis 35 agreement process, such as single argument agreement and DP-internal agreement, and argue that these processes also constitute instances of a general spatial agreement mechanism. Chapter 4 describes the methodology used in this study, which was adapted to the specific sociolinguistic circumstances of LSE as described in section 1.4.2, and gives details of the data. This includes information on the informants, the data collection techniques and how the data were transcribed and analysed. Chapter 5 offers a description of spatial agreement in LSE. The structure of this chapter broadly follows that of chapter 3 so as to provide a comparison between the facts for LSE and what has been described for other sign languages. Thus, descriptions are given for a range of phenomena related to spatial agreement: agreeing verbs, backwards verbs, agreement auxiliaries, and non-manual agreement, as well as single argument agreement and DPinternal agreement. I provide arguments to the effect that all these phenomena involve a basic process of spatial agreement. The chapter also includes a description of the constraints on person/number combinations in agreeing verbs based on a similar study with other sign languages. The comparison makes evident that agreement forms in LSE do not show the same patterns as in other sign languages, but are subject to the same type of constraints (i.e. phonological). Chapter 6 examines the LSE data from a cross-modal typological point of view, drawing on the theoretical framework presented in section 2.2. Firstly, this assessment of the spatial agreement mechanism in LSE takes each element of the agreement process and looks at how the LSE facts compare to the generalizations drawn from typological work on spoken languages. Thus, the controllers, targets, means of exponence, domains, features and conditions that appear in spatial agreement in LSE are held up against what has been described for spoken languages. By and large, LSE fits within the limits of the spoken language data, with the important exception of the features used in spatial agreement in LSE. I claim that person is not a relevant feature for LSE (in contrast to its universal presence in spoken languages) and propose an alternative feature, identity, based on referential identity (developed in chapter 7). The second part of the assessment of the LSE data in chapter 6 involves exploiting the notion of canonicity developed by Corbett (2003b, 2006). Canoncity is defined by a set of criteria that provides a means for evaluating spatial agreement in LSE to see how prototypical or canonical it is as an agreement process. Again, the results show that LSE agreement behaves in a relatively canonical manner. More interestingly, this evaluation highlights 36 Introduction those aspects in which LSE is unusual. Specifically, the optionality of the spatial agreement process is a remarkable characteristic (also taken up in chapter 7). Chapter 7 once more analyses the LSE agreement data but from the point of view of generativist approaches to language structure: minimalist syntax and Optimality Theory. To prepare the terrain, two issues identified in the previous chapter are tackled first: the nature of the identity ϕ-feature and the optionality of the spatial agreement system. Adopting a distributed model of ϕ-features and a feature-sharing theory of agreement, I provide a syntactic analysis of spatial agreement in LSE that can account for the process of location assignment, and verbal agreement for two-place agreeing verbs, for single argument agreement and for agreement auxiliaries. Furthermore, this syntactic account can provide an explanation for the difference between syntactic agreement and a formally similar use of space that leads to different interpretations, namely pragmatic agreement. This chapter also includes a formal analysis of a specific type of agreeing verb that has unusual inflected forms due to a conflict between agreement markers and lexically defined phonological features. Using Optimality Theory, I provide an analysis of the LSE facts that can also be extended to analogous data from another sign language. Chapter 8 concludes this thesis by revisiting the research goals to provide answers to the questions set out in the previous section, and also to assess the extent to which it has been possible to answer these questions. The chapter also includes discussion of the issues that arise from this examination of spatial agreement in LSE, and points the way for future research. Typological approach 43 Before moving on to look at each of these factors, one more stipulation with regard to terminology must be made. Throughout this thesis I make use of the terms “subject” and “object” for ease of exposition. I am aware that these labels make assumptions about the syntactic status of the verb’s arguments, and that there is a need to look beyond such terms to capture the underlying, more primitive notions (McCloskey 2001). Comrie (1989: 70) makes use of a set of labels that obviate pre-empting the question of grammatical relations: S is the argument of an intransitive verb; A is the argument of a transitive verb that correlates most closely with the agent; and P is the argument of a transitive verb that correlates most closely with the patient. While I recognize the value of using such terms when working with a relatively undescribed language, I do not adopt them here (although they are sometimes used when discussing examples from the typological literature). The reason is that the distinctions these labels make do not add anything to the analysis of the LSE data and the challenge of spatial agreement lies in characterizing the use of space rather than the argument that is expressed. 2.2.2. Controllers As Corbett (2006: 35) states, “Controllers are typically nominal in nature.” Nouns and noun phrases often control agreement with adjectives and verbs, respectively, as shown in (3). The adjective agrees in number and gender with the noun personas, and the verb agrees in person and number with the noun phrase las personas ricas: Spanish (3) Las persona-s ric-as trabaj-an poco. DEF.F.PL person(F)-PL rich-F.PL work-3PL little. ‘Rich people work little.’ In the case of verbal agreement, the noun phrase controllers most often fulfil a prototypical semantic role, such as AGENT, THEME or GOAL, as exemplified by (1)-(3). However, other types of argument may be the source of agreement on the verb. In Chicheŵa [Nyanja] (Central Bantu, Niger-Congo), spoken in East Central Africa, the verb agrees in gender with the locative argument in specific constructions. In (4a) the marker kuon the verb ‘come’ marks agreement with the locative argument ‘village’. The fact that the argument is locative, and not another role such as THEME or GOAL (as would be the case in a sentence like ‘The village received those visitors’), is confirmed by the semantics of the verb and the observation that under different word order conditions it agrees with the other argument, a-lendô-wo (‘those visitors’), as in (4b). 44 Theories of agreement Chicheŵa (Bresnan & Kanerva 1989: 2) (4) a. Ku-mu-dzi ku-na-bwér-á a-lendô-wo. G17-G3-village G17.SBJ-REC PST-come-IND G2-visitor-G2.those ‘To the village came those visitors.’ b. A-lendô-wo a-na-bwér-á ku-mu-dzi. G2-visitor-G2.those G2.SBJ-REC PST-come-IND G17-G3-village ‘Those visitors came to the village.’ G=gender class A controller may also be a less typical nominal element than a noun phrase, such as a clause (5a) or an infinitival (5b): Spanish (5) a. Que las modelo-s gan-en tanto parec-e injust-o. COMP DEF.F.PL model(F)-PL earn-3PL.SBJV so_much seem-3SG unfair-M[SG] ‘It seems unfair that models earn so much.’ b. Trabajar no es san-o. work.INF NEG be.3SG healthy-M[SG] ‘Working (lit. to work) is not healthy.’ These controllers are defective since they do not have specific agreement features so the target shows default agreement, which in this case is third person singular on the verbs. This is further shown by the fact that the predicative adjectives in both sentences also agree with the defective controller in the masculine singular default form, despite even the presence of a feminine plural controller within the embedded clause in (5a). Controllers may also be non-overt, as occurs in pro-drop languages such as Pashto (Indo-European), an Indo-Iranian language spoken in Afghanistan, which does not require the subject argument to be explicit (6): Pashto (Neeleman & Szendrői 2007: 672) (6) (Zə) manna xwr-əm. I.NOM apple eat-1SG ‘[I] eat the apple.’ It is also possible for the controller to be completely absent such that it cannot appear, as occurs with verbs that describe natural phenomena, exemplified by the Croatian example (7), in which no overt subject is possible. Typological approach 45 Croatian (Mark Schmalz, personal communication) (7) Mrač-i se. (* Ono rač-i se.) get_dark-3.SG REFL it.N get_dark-3.SG REFL ‘It’s getting dark.’ To summarize, controllers are generally nominal in nature – in the case of verbal agreement they are NPs – and it is possible that they are not overtly expressed. 2.2.3. Targets The examples in the previous section make it clear that the most commonplace targets for agreement are adjectives (example 3) and verbs (all examples). However, there is a series of other elements that mark agreement, such as pronouns, numerals, quantifiers, adpositions and nouns. Before considering these targets, let us look more closely at agreement marking on verbs. 2.2.3.1. Verbs and auxiliaries In nearly all the examples so far, verbal agreement has been marked directly on the verb, but it is also possible for an auxiliary verbal element to bear agreement information. An auxiliary verb may be defined as: an element that in combination with a lexical verb forms a monoclausal verb phrase with some degree of (lexical) semantic bleaching that performs some more or less definable grammatical function. (Anderson 2006: 5) Auxiliary verbs typically express verbal categories of aspect and modality, and may also express tense, negative polarity or voice categories. These categories encode information about the verb semantics, and so do not represent the displaced information that characterizes agreement. However, auxiliaries may also express the features associated with agreement, such as person, number and gender. In (8a) from Burushaski, a language isolate of Northern Pakistan, the auxiliary marks person for the subject argument, and person and number for the possessor of the object argument. Burushaski (Berger 1998: 162, 161, cited in Anderson & Eggert 2001: 240, 242) (8) a. kʰɑkʰɑ ɑy-umuc pʰɑ mée-t-ɑɑ walnut-PL gobble_up 1PL-AUX-2 ‘You gobbled up our walnuts.’ 46 Theories of agreement b. ɑ ɑ ɑ-yúgusɑnc moó-y-ɑ bɑ -ɑ I.GEN 1-daughter.PL 2PL-give-1 AUX-1 ‘I herewith am giving you my daughters.’ In such auxiliary verb constructions, there are a number of ways in which the inflectional material is distributed between the lexical verb and the auxiliary verb. In (8a) the lexical verb pʰɑ (‘gobble up’) is uninflected and the auxiliary mée-t-ɑɑ bears all the inflectional markers; in (8b), in contrast, some information is marked on the lexical verb ‘give’ while the auxiliary still bears an inflectional marker. Although Burushaski shows different ways of distributing inflectional material between the verb and auxiliary, many languages consistently use one pattern. All information may appear on the auxiliary; an example of such an AUX-headed language is Iatmul (Sepik), as can be seen in example (9), in which subject marking appears on the auxiliary. Iatmul (Foley 1986: 144, cited in Anderson 2006: 24) (9) klə-kə lɨ-kə-wɨn get-DEP AUX-PRES-1SG ‘I am getting it.’ DEP=dependent Alternatively, it may be the lexical rather than the auxiliary verb that bears the person inflection; this is the case for Mödö (Nilo-Saharan), shown in example (10), which marks for subject on the lexical verb ‘rescue’. Mödö (Persson & Persson 1991: 19, cited in Anderson 2006: 24) (10) tí mók nyì yí FUT 1:rescue you ‘I will rescue you.’ Another possibility is that both elements are inflected: example (11) shows how in Gorum (Austro-Asiatic) both the lexical verb ‘eat’ and the auxiliary bear marking for the first person subject. Gorum (Aze 1973: 279, cited in Anderson 2006: 25) (11) miŋ ne-gɑʔ-ru ne-lɑʔ-ru I 1-eat-PST 1-AUX-PST ‘I ate vigorously.’ The inflectional material may be divided between the lexical and auxiliary verbs such that each element bears different information; Jakaltek (Mayan) is Typological approach 47 a split language of this type and example (12) shows how the subject is marked on the lexical verb ila (‘see’) while the object is marked on the auxiliary. Jakaltek (Craig 1977: 60, cited in Anderson 2006: 25) (12) -ɑch w-ila COMPL-ABS2 ERG1-see ‘I saw you.’ Anderson also identifies a fifth category of auxiliary verb constructions, which he calls the split/doubled type. In languages of this type, the information is marked on both the lexical and the auxiliary verbs, but incompletely so on one of the two elements. Burushaski has split/doubled auxiliary verb constructions, as can be seen in (8b): the subject is marked on both elements, but the object is marked on the lexical verb alone. The different types of auxiliary verb constructions are summarized in table 1.1. LEX AUX AUX-headed type – + LEX-headed type + – Doubled-type + + Split type +i/‒ j – i/+j Split/doubled type +i/+ j – i/+j – i/+j + i/+j Table 1.1. A typology of auxiliary verb constructions showing the possibilities for the distribution of the inflectional material between the lexical verb (LEX) and the auxiliary verb (AUX). (Adapted from Anderson 2006: 24-27.) The division of labour between the lexical and the auxiliary verb in the split and split/doubled types varies from language to language, and there are different groupings that contrast negation, TAM (tense, aspect and modality) and subject/object marking. Leaving aside negation, the inflection on the lexical verb and auxiliary verb tends to differentiate between subject on the one hand and object on the other, or between TAM on the one hand and subject/object on the other. As we shall see in chapters 5 and 6, this second pattern is comparable to how agreement auxiliaries in LSE operate. For the moment we return to the different types of targets that can mark agreement. 2.2.3.2. Other targets of agreement In addition to verbal elements, agreement marking may appear on other categories of word. This section looks at those that are relevant to agreement 48 Theories of agreement in LSE. As we shall see in chapter 3, the spatial marking that appears on sign language verbs is closely related to the use of space also present in pronouns; additionally, spatial marking in LSE may also appear on numerals, quantifiers, adpositions and nouns, so agreement on these categories in spoken languages is described here. Pronouns often mark the person, number or gender of their antecedent, or a combination of these features. In the case of Tamil (Dravidian), the pronoun marks person, number and honorific status. In example (13), the pronoun avaru matches the third person singular honorific antecedent mutal mantiri (‘Chief minister’). Tamil (Asher 1985: 4-7, cited in Brown et al. 2002) (13) A utta vaaram mutal mantiri namma uur-ukku var-ra-aru. next week first minister our village-DAT come-PRS-3SG.HON avaru a utta vaaram va-ruva-aru PRO.3SG.HON next week come-FUT-3SG.HON ‘The Chief Minister is coming to our village next week. He will come next week.’ PRO=pronoun This would not be treated as agreement under many models, since the pronoun does not have to be within a certain distance of the antecedent with which it agrees (see section 2.2.4, on domains). Such a model is binding theory (Chomsky 1981), in which a pronoun is taken to be subject to specific restrictions regarding its relation and relative position with respect to the antecedent. Condition B of binding theory (“a pronominal is free in its binding domain”) basically requires that the antecedent does not appear in a syntactic position in which it controls the pronoun, and thus no relation holds. However, pronouns display the same agreement features (such as number, person and gender) that typically show up on agreement markers. Barlow (1999: 200) claims that agreement and antecedent-anaphora relations make use of the same underlying mechanisms, even though there may be more going on in the case of anaphoric reference (concerning distribution and control). Corbett (2006: 228-30) supports this view by pointing out that there is no obvious or logical point at which agreement phenomena can be qualitatively categorized into different types (e.g. local versus anaphoric agreement), and argues for a unified model of agreement. The question of relation or proximity required between a target and its controller will be looked at in the section on domains, and returned to in the overview of agreement in the Minimalist framework in section 2.3.3. Typological approach 49 Another element that can mark agreement is numerals. It is common for the numeral ‘one’, which is often closely related to the indefinite article, to show gender agreement with the noun controller, but higher numerals may also show variance in some languages. In Catalan, the numeral ‘two’ marks the gender of the controller noun, as can be seen by the distinction between dugues with the feminine noun ampolla (‘bottle’) and dos with the masculine noun bidon (‘can’) in the examples in (14). Catalan (14) a. dugues ampolle-s d’ aiga two.F bottle(F)-PL of water(F) ‘two bottles of water’ b. dos bidon-s d’ aiga two.M can(M)-PL of water(F) ‘two cans of water’ This forms part of a general cross-linguistic pattern: the lower the numeral, the more likely it is to be a target of agreement. Conversely, the higher the numeral, the less likely it is to show agreement, and higher numerals show more irregular agreeing patterns: in Russian, the numeral ‘two’ distinguishes the feminine from the masculine and neuter genders, while ‘three’ and ‘four’ do not distinguish gender but do agree according to animacy (Corbett 1991, 1993). Quantifiers and question words may also show agreement with a noun controller. This occurs in Turkana (Nilo-Saharan), spoken in Kenya, shown in the examples in (15) by the alternation of the form of the word ‘which’ depending on the gender of the controller noun. Turkana (Dimmendaal 1983a: 433-434, cited in Brown et al. 2002) (15) a. e-kìle a-li` M.SG-man M.SG.which ‘Which man is it?’ b. ɪ-k kʊ a-ni` N.SG-child N.SG.which ‘Which child is it?’ Adpositions may also show agreement and typically agree with the noun they govern, as occurs in the case of many modern Indic languages, some of the Iwaidjan languages (of Northern Australia) and the Celtic languages (Corbett 2006: 46). Example (16) shows agreement in person and number between the 50 Theories of agreement Welsh preposition am (‘about’) and the noun it governs (only in the case where the noun is pronominal). Welsh (adapted from Sadler 2003, example 7) (16) a. Roedd Wyn yn siarad amdanat (ti). was.3SG Wyn PROG speak about.2SG 2SG ‘Wyn was talking about you.’ b. Roedd Wyn yn siarad amdanom (ni). was.3SG Wyn PROG speak about.1PL 1PL ‘Wyn was talking about us.’ The last category we shall look at that can mark agreement are nouns themselves. As we have seen, nouns typically control agreement on another element such as a verb or adjective, but they may also be marked to show agreement. Noun targets usually agree with some other noun, and often in the context of a possessive construction, in which the possessor agrees with the possessum, or vice versa. In Palauan (Austronesian), spoken in various islands of the Western Pacific, the possessum agrees with the possessor (17). Palauan (Josephs 1975: 66-68, glossed following Potet 1992 and Tiberius 2002) (17) a. a bli-l a Droteo PM house-3SG.POSS PM Droteo[SG] ‘Droteo’s house’ b. urer-ir a re-dil work-3PL.POSS PM PL-woman ‘women’s work’ PM=phrase marker It is common for nouns to agree in number when they are predicative, so as to avoid a semantic mismatch. However, it should be kept in mind that there are languages that do not require a noun predicate to agree in number, especially when the subject is inanimate or non-human. This is the case for Hungarian, which shows number agreement in (18a), but not in (18b) for the non-human subjects. Hungarian (Hall 1944, cited in Brown et al. 2002) (18) a. Molnár és Kostolányi Magyar iró-k Molnar and Kostolányi Hungarian writer-PL ‘Molnar and Kostolanyi are Hungarian writers.’ Typological approach 51 b. a tehén és a ló leg-hasznosabb háziállat ART cow and ART horse most-useful domestic_animal ‘The cow and the horse are the most useful domestic animals.’ So far we have seen which types of elements may mark agreement, namely verbs (including auxiliary verbs), adjectives, pronouns, numerals, quantifiers, adpositions and nouns. Now let us turn to the matter of how these targets mark agreement. 2.2.3.3. Means of exponence The most common way of manifesting agreement is by means of affixes. This is what we have seen in the examples so far: mainly suffixes, but also prefixes in the case of gender marking in Chicheŵa (4) and infixes in the case of number agreement with the (absolutive) object in Basque (2). A particular type of affixal agreement has received some attention in the literature: alliterative agreement, common in Bantu and other Niger Congo languages. Alliterative agreement involves the presence of a marker on the controller and the same marker is used for agreement targets. In the Swahili (Central Bantu, Niger-Congo) example in (19), the singular marker for gender 7/8, ki-, is repeated on the adjective, the numeral and the verb that agree with the subject. Swahili (Welmers 1973: 171, cited in Corbett 2006: 87) (19) ki-kapu ki-kubwa ki-moja ki-li-anguka SG-basket(G7/8) G7-large G7-one G7-PST-fall ‘One large basket fell.’ The interesting cases arise when the controller lacks a gender marker, which is a common situation for loan words. With no marker on the controller there is nothing available to be re-used to mark agreement on the targets. In this case, two options are available: in some cases a default marker is used, but in others the marker copies part of the phonological form of the controller. An example of this type of radical or literal alliterative agreement is shown in (20) for Bainouk (Atlantic, Niger-Congo): (20a) shows typical alliterative agreement for a noun that bears a gender marker; in (20b), on the other hand, the noun has no gender marker and the numeral uses the second strategy by copying the first consonant and vowel (CV) to mark agreement. Such radical alliterative agreement has been claimed to exist in other languages such as Arapesh (Toricelli), Wolof (Niger-Congo) and Landuma (Landoma, NigerCongo), but evidence for a robust mechanism is scant (Corbett 2006: 90). 52 Theories of agreement Bainouk (Sauvageot 1967, cited in Aronoff, Meir & Sandler 2005: 320) (20) a. gu-s l gu-fɛr G7-tunic G7-white ‘white tunic’ b. kata:m-ã ka-nak-ã river-PL CV-two-PL ‘two rivers’ The importance of radical alliterative agreement lies in its undermining of the principle that syntax is phonology-free, since the (syntactic) agreement process needs to know about the phonological form of the controller. I will not enter into this debate, but note that the phenomena is relevant to sign language agreement as parallels have been drawn between radical alliterative agreement and agreement in sign languages (Aronoff, Meir & Sandler 2005). Some verbal affixes fall into a grey area with respect to agreement, and there is a division of opinion as to whether or not they constitute a manifestation of agreement proper. These are bound markers for person, number and gender that are treated as pronouns attached to the verb rather than agreement markers. The distinction is often a difficult one to make, but is of consequence since from a syntactic point of view a pronoun is an argument of a verb, whereas a verb marked for agreement has independent arguments (though they may not be overt, as in the case of pro-drop languages). Corbett (2006: 110) points out that the difference is also important if agreement is restricted to the clausal level, since pronouns may be indexed (on this view they do not “agree”) with antecedents outside the immediate clause, whereas agreement markers must stand in a local relation to their controllers. The Australian language Bininj Gun-Wok [Gunwinggu] (Australian) makes use of pronominal affixes, shown by the prefix gabanon the verb (‘scold’) in example (21). Bininj Gun-Wok: Gun-djeihmi dialect (Evans 1999: 266, cited in Corbett 2006: 104) (21) al-ege daluk gaban-du-ng bedda F-DEM woman 3SG>3PL-scold-NPST them ‘That woman is scolding them.’ These pronominal affixes are hybrid elements as they are morphologically bound to the verb, yet at the same time they are arguments of the verb and referentially they function much like independent pronouns (Mithun 2003). In this sense they fall between typical agreement markers and free pronouns, Typological approach 59 b. JON IXx MIREN IXy PROBLEM xEXPLAINy ‘Jon explained the problem to Miren.’ We have established that the typical domain for (verbal) agreement is between a verb and its arguments, but there are attested cases of agreement where the verb is controlled by an element that is not one of its arguments. The first case is possessor agreement, in which the verb agrees with the possessor of an argument rather than the argument itself. 4 We have already seen an example of this in Burushaski, repeated here as (28). The auxiliary verb bears the first person marker méeto indicate who the walnuts belong to, rather than indicating agreement with the walnuts themselves. Burushaski (Berger 1998: 162, 161, cited in Anderson & Eggert 2001: 240, 242) (28) kʰɑkʰɑ ɑy-umuc pʰɑ mée-t-ɑɑ walnut-pl gobble_up 1PL-AUX-2 ‘You gobbled up our walnuts.’ This agreement between the verb and the possessor of an argument occurs in several languages, such as Maithili (Indo-European), Banawá (Jamamadí, Arauan), Tabasaran (Daghestian) and Fox [Meskwaki] (Algic) (Corbett 2006: 61; Anderson 1997: 234). For this type of structure, it has been claimed that the possessor undergoes movement to a position typically occupied by verbal arguments (“possessor raising”), thus providing the syntactic justification for the manifestation of the agreement relation (Landau 1999). However, such raising analyses are ruled out for certain languages, such as Maithili, on the grounds of word order considerations: in spite of relatively free word order possibilities in the language, the possessor cannot be extracted from the major constituent of which it forms part. This shows that syntactically the possessor must be part of the containing possessum NP rather than a separate major constituent in the clause and thus it is hard to know what syntactic relation between the possessor and the verb constitutes the agreement domain (Comrie 2003: 335). Conversely, the agreement controller does appear to occupy an argument position for the verb in the copying-to-object formations described by Anderson (1997: 231-233). In these constructions the argument of an embedded clause becomes the object of the matrix verb. In the example in 4 Possessor agreement between a verb and the possessor of one its arguments is to be distinguished from agreement between possessor and possessum, as mentioned in section 2.2.3.2. Recall that this discussion of domain focuses on verbal agreement, and as such ‘possessor agreement’ here refers to agreement with a possessor on the verb (not on a possessum). 60 Theories of agreement (29a) from the North American language Fox, the verb ‘want’ could take a clausal object (which would be treated as an inanimate noun) but inflects for a third person (animate) object. Anderson claims that this is a copying mechanism rather than some sort of raising effect since the referent is signalled twice and both clauses are potentially free-standing. From the point of view of agreement domains, there is nothing too unusual going on here: once the argument is copied to the object position in the matrix clause, the verb marks agreement with it. However, Anderson reports cases in which the element that triggers agreement on the matrix verb has little to do with the embedded verb’s argument structure. In (29b), the embedded first person topic is what controls agreement on the matrix verb. Fox [Meskwaki] (adapted from Anderson 1997: 232, 233) (29) a. net-ɑ ɑ:wɑ:n -ɑ:-wɑ=koh(i) wi:h=ne:w-ɑ 1-want-DTS(1 >)-3.IIND=you_know FUT=see-1>3.AOR Lit. I want him (that) I will see him. ‘I do want to see him.’ b. ne-kehke:nem-ek(w)-wɑ ni:nɑ e:h=pwɑ:w -ke:ko:hi-ɑšeno-niki l-know-INV-3(>1).IIND 1.TOP AOR=not-anything-disappear-INAN.OBV.AOR Lit. He knows me that as for me nothing is missing. ‘He knows (that) as for me nothing is missing.’ DTS=direct theme sign, INV=inverse, IIND=independent indicative, AOR=aorist, INAN=inanimate, OBV=obviative (29b) could perhaps be treated as a combination of possessor raising in the embedded clause and copying-to-object into the matrix clause. Semantically, the possessor agreement construction is reminiscent of the dative of interest common in Romance languages but rather than occupying an oblique argument position as occurs in (30), the referent in the possessor agreement and copying-to-object formations is marked as an immediate argument on the verb. Spanish (30) Este niño no me/te/le com-e nada. This child NEG me/you/him/her.DAT eat-3SG nothing ‘This childi will eat nothing (and I’m/you’re/(s)hej’s affected).’ Together with possessive agreement and other data, examples like (29b) are used by Anderson to argue that agreement in Fox is conditioned by discoursedriven considerations to do with highlighting animate referents salient to the discourse. Initially, we had described these agreement-with-non-argument Typological approach 61 phenomena – possessor agreement and copying-to-object formations – as being clause internal, but Anderson’s line of reasoning pushes the limits back beyond the clause (and the sentence) to the level of discourse. 5 Under this view, agreement is (or, put more cautiously, in some languages may be) sensitive to effects that are not contained within the morphosyntactic domain and cannot be described in terms of (syntactic) structural relations. A more radical view of “agreement as a (purely) discourse phenomenon” is offered by Barlow (1999), who suggests that the relationship between controllers and targets cannot be reduced to feature identity and is better captured as a merging of interpretations associated with discourse referents. The characterization of agreement in terms of discourse considerations sits in stark contrast with the minimalist tack, which considers the domain of agreement to be within narrow syntactic structure (see section 2.3.3). The last unusual agreement effect that we shall look at is a clear case of agreement beyond the clause, regardless of the role given to discourse considerations. Long distance agreement has been attested for various languages such as Godoberi (North Caucasian) (Haspelmath 1999), Basque (Isolate) (Etxepare 2005), and Lokaa (Cross River, Niger-Congo) (Baker 2008). Example (31) comes from the Daghestanian language Tsez [Dido] (North Caucasian), showing how the matrix verb ‘know’ agrees in gender with the object of the embedded clause, ‘bread’. Tsez [Dido] (Polinsky & Potsdam 2001: 584) (31) enir už magalu b c’ruł b-iyxo mother boy bread(G3).ABS ate G3-know ‘The mother knows the boy ate the bread.’ It can be shown that movement in Tsez is strictly clause internal and there is no way for an argument to move outside of its clause. This means that the agreement between the verb and the embedded element must be across a clause boundary. Polinsky & Potsdam (2001: 641) conclude that this makes long distance agreement “problematic for theories of agreement that either explicitly stipulate or axiomatically derive the claim that all agreement relationships are clause bounded.” To summarize this section, the domain of agreement is the relationship that holds between the two elements involved in the agreement configuration, 5 Another phenomenon which recommends the importance of taking into account (animate) referents salient to the discourse is allocutive agreement in Basque, in which the verb agrees with the addressee in person and gender even though the referent is not an argument selected by the verb (Oyharçabal 1993). 62 Theories of agreement the controller and the target, and typically this is defined in terms of grammatical function (“the verb agrees with its subject”). We have looked at various agreement phenomena that resist any attempt to identify a straightforward connection between the controller and the target: in possessor agreement, the verb agrees with the possessor of its logical object; copying-toobject formations show the verb agreeing with a discourse salient referent (which plays no part in the verb’s argument structure); and long distance agreement allows the matrix verb to agree with the argument of a different verb in an embedded clause. One reaction to these unusual agreement effects is to say that agreement is sensitive to discourse factors or, more radically, that it operates entirely at the level of discourse. An alternative strategy, taken up by the Minimalist Program, is to characterize the domain of agreement in purely structural terms, limiting the description of the relation between controller and target to syntactic configurations. 2.2.5. Features and values Features are the information from the controller that is marked on the target in the agreement process. More precisely, a feature is the type of information that is marked and the specific information shown is the value; for example, a verb may agree in number (the feature) with its object and in a given case that may be dual (the value). A feature has a set of possible values that varies from language to language: in the case of Northern Ostyak [Khanty] (Uralic), number may be singular, dual or plural, as can be seen from the different markers on the verb we:l (´kill´) in (32). Northern Ostyak [Khanty] (Nikolaeva 1999: 334) (32) a. ma tam kalaŋ we:l-s-Ø-e:m I this reindeer kill-TENSE-SG-1SG.SBJ ‘I killed this reindeer.’ b. ma tam kalaŋ we:l-sə-ŋil-am I these reindeer kill-TENSE-DU-1SG.SBJ ‘I killed these (two) reindeer.’ c. ma tam kalaŋ we:l-sə-l-am I these reindeer kill-TENSE-PL-1SG.SBJ ‘I killed these reindeer.’ The main features for agreement, the ϕ-features, are gender, number and person. There are other features that may be considered in the agreement process, such as case and respect. Before looking at each of these features and Typological approach 63 their sets of possible values, I wish to look at the general properties of features themselves. A feature may be categorised in terms of whether or not it is required by the syntactic context, in much the same way that inflection may be considered inherent or contextual (cf. Booij 1996). Applying this distinction to features, a contextual feature is one that is required by the syntactic context, while an inherent feature is not, although it may be relevant to the syntactic system. This property of a feature depends upon where the feature appears: gender is inherent for nouns, but contextual for adjectives (Corbett 2006: 123). A further distinction (due to Zwicky 1992, cited in Corbett 2006: 124) may be drawn with respect to how a feature relates to semantics. Direct features express intrinsic content and are associated with prototypical semantics (number with numerosity, gender with a classification of objects, etc.). Indirect features, on the other hand, express meanings indirectly, by means of grammatical relations (nominative case is associated with the grammatical relation of subject, which in turn is associated with the semantic role of AGENT). These two properties coincide closely but are not the same: the first relates to syntax (and as such depends upon where the feature appears), while the second is to do with semantics. It is possible for a feature to be inherent and indirect and this is of importance for the theoretical apparatus of Minimalism (cf. uninterpretable features in section 2.3.3). Another important property of features is that they represent a certain level of abstraction. A feature’s values act as markers that categorize nominal elements as belonging to a given category, such as plural in number, or masculine in gender. These values have a semantic basis and serve as a means to carve up the world of linguistic elements into different types, which is made use of by the grammatical system. As such, many different items may share the same ϕ-features (and values). This means that ϕ-features do not uniquely individuate specific items, and they are to be contrasted with indices, which are specific labels for a single item. The distinction will be an important one when we come to look at the features at work in agreement in LSE. We now turn to the different features that play a role in agreement. 2.2.5.1. Gender Gender is an inherent feature of the noun and categorizes it according to some sort of semantically based taxonomy. The better known gender systems have two or three values: masculine, feminine and neuter. However, other languages, such as the Bantu languages, have more involved gender systems that normally distinguish between seven and ten genders (or classes, as they are known in the Bantuist tradition). In the extreme, Nigerian Fula (Niger- 64 Theories of agreement Congo) has around twenty genders, depending on the dialect (Corbett 2008a). Gender is an unusual feature in that it may or may not figure in a language’s grammar: in a survey of 257 languages for the World Atlas of Language Structures (WALS – Haspelmath, Dryer, Gil & Comrie 2008), Corbett (2008a) finds that over half (145) have no gender system. Unsurprisingly, when a gender system is present, the more distinctions it makes, the less common it is. Gender systems may be based on purely semantic criteria or a combination of semantic and formal criteria. In the first case, the categories established by the system are directly related to the meaning of the members in each category, as is the case for Kannada (Dravidian) for which all male humans are masculine gender, all female humans are feminine, and everything else is neuter (Corbett 2008c). Alternatively, the gender system may have a semantic base supplemented by other criteria that result in categories with a mixed set of members that do not seem to form a natural class of any sort. The additional assignment rules often take into account the form of each noun, and this may be done on the basis of phonology or morphology. The same WALS survey of gender systems found a roughly equal split between strictly semantic and mixed semantic/formal gender systems (Corbett 2008c). A final consideration for gender systems is the underlying distinction of the semantic criteria for assigning gender. The majority of languages that have a gender system in the WALS survey (84 of 112) applies a sex-based categorization and the remaining languages make use of animacy as the basic differentiating factor. In the most limiting case, animacy is restricted to humans but it may also be extended to animals and spirits or trees. Many of the languages that make use of an animacy-based gender system are from the Niger-Congo and Algonquian families, but this type of language is also represented all over the world (Corbett 2008b). In summary, gender is an inherent feature of nouns that stems from a semantic classification based on either sex or animacy, which may be obscured by additional formal criteria for gender assignment. Not all languages have gender systems, and the extent to which gender participates in agreement processes may vary from language to language: gender agreement is generally limited to the nominal domain, but may also play a role in verbal agreement. The following section looks at the feature of number. Typological approach 65 2.2.5.2. Number Number is an inherent feature of nouns 6 and its value is normally marked on the noun (e.g. a plural marker), but some nouns may be lexically specified for number. Most commonly, the number feature distinguishes two values: singular and plural. The plural category may be further split into more specific values. Dual marks two and only two referents in Upper Sorbian (Indo-European) (Corbett 2000: 20) and Hopi (Uto-Aztecan) (Moravcsik 1978: 347), thus restricting the plural value to three or more. Trial marks three and only three referents and appears in languages that have a dual form, such as Larike [Larike-Waksishu] (Central Malayo-Polinesian, Austronesian) (Corbett 2000: 21). The paucal is used to refer to a small number of distinct referents: in Bayso (Afro-Asiatic) the paucal refers to between two and six individuals (Corbett 2000: 22). The paucal and the general plural may also be divided into lesser and greater categories, with the result that languages may have up to five different values for the number feature: Mele-Fila (Central-Eastern Oceanic, Austronesian) distinguishes between singular, dual, paucal, plural and greater plural (Corbett 2000: 42). In addition to these number distinctions, some languages have a general value that is outside or beyond the number system. A language may have a specific form that is neutral with respect to number. In the Fouta Jalon dialect of Fula (Niger Congo), the word toti may refer to one or several toads, and contrasts with the forms for the singular, totii-ru (‘a toad’), and plural, totii-ji (‘toads’). Many languages have a general meaning but rather than use a separate form, this is achieved by means of one of the forms for more restricted number meaning. Thus, in Turkish ev can mean ‘house’ or ‘houses’, while the plural form evler always means ‘houses’ (Corbett 2000: 10-14). This notion of general number will turn out to be relevant when interpreting the LSE data, especially given the apparent optionality of number marking. Number as a nominal feature needs to be distinguished from verbal number. We are used to thinking of number as being a feature that a verb agrees with (that is, is expressed as a contextual feature on the verb as a target), but a verb may have number as an inherent feature. Verbal number reflects the event semantics of the verb and indicates whether an action is performed several times or at several places (event number), or whether it affects or involves several participants (participant number). In this sense, verbal number may overlap with both aspect (iterative and distributive) and 6 The notion that number (and gender) is inherent to a noun actually depends on how this information is represented in the syntactic structure. I return to this issue in section 7.1.1 when re-examining the location of ϕ-features. 66 Theories of agreement nominal number as reflected in agreement (Veselinova 2008; Corbett 2000: 256). However, it is often possible to distinguish verbal number as a grammatical category that is marked on the verb. In Georgian (Kartvelian), the form of the verb stem ‘sit’ marks the plurality of the action (one person sits vs. several people sit). The verb also bears an agreement marker that indicates the number of the controller argument. The contrast between (33a) and (b) highlights the different verbal forms associated with a singular subject argument/single event (i.e. singular verbal number) and with a plural subject argument/multiple event (i.e. plural verbal number). However, it is possible to distinguish between nominal (argument) number and verbal number due to the fact that in Georgian numerals require a singular noun and control singular agreement on the verb. Thus, in (33c), with the numerically quantified subject ‘my three friends’, the verb is marked to show agreement with a singular subject by the affix -a, similarly to (33a), but also contains the affix -sxd-, similarly to (33b), to mark plural verbal number. Georgian (Aronson 1982: 243, 406-7, cited in Corbett 2000: 254) (33) a. ivane še-mo-vid-a da da-ǰd-a John PRV-PRV-enter-AOR.3.SG and PRV-sit.SG-AOR.3.SG ‘John entered and sat down.’ b. čem-i mšobl-eb-i še-mo-vid-nen da da-sxd-nen my-AGR parent-PL-NOM PRV-PRV-enter-AOR.3.PL and PRV-sit.PL-AOR.3.PL ‘My parents entered and sat down.’ c. čem-i sam-i megobar-i še-mo-vid-nen da my-AGR three-AGR friend.SG-NOM PRV-PRV-enter-AOR.3.PL and da-sxd-a PRV-sit.PL-AOR.3.SG ‘My three friends entered and sat down.’ PRV=preverbal Verbal number is marked on the verb and is an inherent feature of the verb. As such, it does not represent a case of agreement since there is no displaced information. However, it is relevant to agreement because, as we have seen, the verb may also carry agreement markers that reflect the number feature value of one (or various) of its arguments. This is true for Georgian, and it was relatively straightforward to distinguish the two phenomena. However, it is not always so easy to differentiate between verbal number and agreement markers. Durie (1986: 357-62) provides the following diagnostics: Typological approach 67 i) verbal number operates on an ergative basis, reflecting the number of the most directly affected participant, which is the subject of intransitive sentences (S) or the object of transitive sentences (P), and this may contrast with other marking on the verb (e.g. subject marking, which agrees with S and A). ii) verbal number may mark different values to those marked by agreement, especially when verbal agreement is restricted by some condition (such as singular agreement for numeral phrases in Georgian). iii) verbal number may have a different set of values to nominal number; although rare, it is possible for verbal number to include a value (such as dual) that is not marked by nominal number in the same language, or vice versa. iv) verbal number is retained in contexts where agreement is absent, namely non-finite forms that lack agreement morphology such as control constructions, imperatives and attributive usage. v) verbal number is preserved in derivational word formation, but agreement inflection for (nominal) number is not. These differences between verbal number and nominal number marked on the verb serve to identify how much of the inflectional material on a verb is due to agreement (and conversely, how much is marking inherent features of the verb). This will be useful for delimiting verbal agreement in LSE. Number differs from gender due to the fact that it is held to be universally present in all languages: Universal #42 proposed by Greenberg (1963: 113) states “All languages have pronominal categories involving at least three persons and two numbers.” This is not quite true. Firstly, there are languages that make no grammatical distinction in number. Corbett (2000: 5051) mentions Kawi (Austronesian), Classical Chinese (Sino-Tibetan) and Pirahã (Mura) as examples of languages that have no plural nouns or pronouns (though semantic number may be expressed by means of conjunctions and quantifiers). Example (34) shows how the third person pronoun is used indistinctly in Pirahã for singular or plural reference. Pirahã (Everett 1986: 282, cited in Corbett 2000: 51) (34) hiapióxio soxóá xo-ó-xio 3 already jungle-LOC-DIR ‘He already went to the jungle.’ or ‘They already went to the jungle.’ DIR=directional 68 Theories of agreement A second problem with Greenberg’s universal is that it is couched in terms of pronominal categories and the question of what counts as a pronominal crosslinguistically is a thorny issue. This is closely connected to the category of person so we shall look more closely at this matter in the following section. 2.2.5.3. Person Person is a feature inherent to pronominal elements rather than nouns, which are taken to be third person by default. 7 According to Greenberg’s Universal #42, person is a universal feature of the pronominal category and always distinguishes between three different values: first (the speaker), second (the addressee) and third (neither speaker nor addressee). Since the grammatical category of person is closely linked to the category of personal pronoun, the definition of what counts as a personal pronoun is central to assessing the universality of person. Cysouw (1997) claims that some languages, such as West-Greenlandic Inuit (Eskimo-Aleut), lack a third person pronoun, making use instead of a demonstrative. Cysouw’s definition of a third person pronoun is as an intersubjective deictic: “an item that can be used by all speech-act participants alike to refer to something” (1997: 9). Since demonstratives are not intersubjective (‘this’ for me may be ‘that’ for you, while ‘he’ remains constant for me and you) he concludes that they are not pronominal and thus that third person pronouns are not a universal category. In a similar vein, if pronouns are defined as a morphosyntactic category, there are languages such as Thai (Tai-Kadai) and Japanese (Japonic) whose person markers behave more like nominals than pronominals, with the result that it is not clear that all languages have a pronominal category (Siewierska 2004: 9). There are two responses to these claims. The first is to use an alternative definition of pronouns in terms of referential role and functions which focuses on their referential deficiency (to distinguish them from nouns) and anaphoric 7 However, some languages do allow non-third person agreement with a nominal phrase. Spanish is often cited in this respect because of examples like the following: (i) a. ¿El grupo enter-o v-áis? DEF.M.SG group(M.SG) whole-M.SG go-2PL ‘Is the whole group (of you) going?’ b. Los marica-s abund-amos en est-e campo. DEF.M.PL poof(M)-PL abound-1PL in this-M field(M) ‘(Us) poofs are in abundance in this field.’ Typological approach 75 shortcomings, two of which I will address. Firstly, from a practical point of view, establishing the norm would be a huge undertaking as it would involve taking stock of the agreement systems of all the world’s languages. Secondly, a statistical approach could also run into problems due to the levelling nature of averages. To give a facile illustration, take gender as a case in point. Using the figures for languages with different numbers of gender values from the WALS sample of 257 languages (Corbett 2008a), a rough calculation gives an average of around 2 genders. Yet the most common category is for a language not to mark gender at all. Equally, a simple statistical average fails to capture the interaction between different factors: continuing with the gender example, the fact that non-sex-based gender systems tend to have many more values than sex-based ones would be lost in an averaging process. Obviously, these errors could be overcome by improved statistics (in these cases using the mode instead of the mean, and more complex variance statistics), but there remains an underlying problem of failing to capture the full extent of the agreement phenomena. Establishing a statistical norm fails to delimit the extent of the phenomenon and only gives us an image of the most populated part of the agreement terrain rather than the peripheries. An alternative approach, based on the notion of canonicity, examines the different ways in which agreement can vary and stipulates criteria based on these variables. The criteria lay out the different options for agreement systems and thus provide a mapping of the theoretical space of possibilities. For each variable a canonical value is designated in accordance with general principles that are deemed to characterize (canonical) agreement. This means that the most canonical system is the one that best conforms to the general principles and is not necessarily the most commonly occurring system among the world’s languages. Each of the criteria provides a parameter to evaluate a given agreement system against the prototypical agreement system. Corbett (2003b, 2006: 10-27) develops a canonical approach that I will adopt here. I limit myself to listing Corbett’s principles and criteria, providing explanation only where the terminology demands it. Readers interested in the motivation and justification for Corbett’s choices should refer to his work. The general principles of canonical agreement are as follows: Principle I: Canonical agreement is redundant rather than informative. Principle II: Canonical agreement is syntactically simple. Principle III: The closer the expression of agreement is to canonical (i.e. affixal) inflectional morphology, the more canonical it is as agreement. 76 Theories of agreement These principles dictate what the more canonical value is for the different parameters, which are set out in table 1.3. C-1: controller is present > controller is absent C-2: controller has overt expression of agreement features > controller has covert expression of agreement features C-3: consistent controller (all targets take the same value for a given feature) > hybrid controller (targets take different values for a given feature) C-4: controller’s part of speech is irrelevant > controller’s part of speech is relevant C-5: marking is bound > marking is free C-6: marking is obligatory > marking is optional C-7: marking is regular (affixal) > marking is suppletive C-8: marking is alliterative (marker on all targets is the same and identical to formant on controller) > marking is opaque (marker changes from target to target and is not identical to formant on controller) C-9: marking is productive (applies to all members of a category) > marking is sporadic (only appears on some members of a category) C-10: target always agrees > target agrees only when controller is absent C-11: target agrees with single controller > target agrees with more than one controller C-12: target has no choice of controller > target has choice of controller C-13: target’s part of speech is irrelevant > target’s part of speech is relevant C-14: domain is asymmetric > domain is symmetric C-15: domain is local > domain is non-local C-16: domain is one of a set > single domain C-17: feature is lexical > feature is non-lexical C-18: features have matching values > feature values do not match C-19: no choice of feature value > choice of feature value C-20 no conditions > conditions Table 1.3 Criteria for canonical agreement. The symbol > means “is more canonical than”. Adapted from Corbett (2006: 10-27). These criteria provide a gauge of how canonical agreement in a given language is, and may be applied to the agreement system of a language as a whole, or to specific aspects of agreement for that language. Thus, for a given language verb agreement may be strongly canonical while DP-internal agreement is less so. I shall apply this notion to agreement in LSE to give us an idea of whether or not LSE has an agreement mechanism and, more Typological approach 77 generally, whether what has been treated as agreement in the sign language literature is justifiably labelled as such. 2.2.8. Summary In this section, we have looked at agreement from a typological point of view, surveying the phenomenon from the perspective of the diversity of its manifestations across the world's (spoken) languages. In order to accommodate this variation, no rigid definition of agreement is stipulated, but rather a terminological framework that can describe the different types of agreement that are attested. This descriptive approach identifies the different elements that enter into play in the agreement relationship, and we have examined each of these in turn: controllers, targets, features, domains and conditions. Controllers, the elements agreed with, are generally nominal elements (in the case of verbal agreement nominal phrases), which may or may not be overtly present. Targets are the elements that agree with a controller, and carry some sort of marking that shows the agreement. There is greater heterogeneity among targets and we have seen that verbs (both lexical and auxiliary verbs), adjectives, pronouns, numerals, quantifiers, adpositions and nouns may be targets. There is also a variety of means by which agreement is marked on the target including inflection, clitics and multiple exponence. Two types of agreement marking will be especially relevant to LSE. The first is alliterative agreement, found in many Bantu languages, which involves the apparition on the target of a formant (such as a gender-marking prefix) already present on the controller, and the more exceptional case of radical alliterative agreement, which involves copying phonological material from the controller onto the target (often because no formant is available). The second type of marking is pronominal affixes, which represent a grey area between agreement markers and free pronouns. The domain is the context in which the agreement relation holds between the controller and the target, and is generally some sort of local grammatical relation or syntactic configuration. Delimiting the agreement domain is for many the defining factor for what counts as agreement and what does not. From the typological perspective of this section, we have looked at the variety of the phenomenon and this has included instances of “badly behaved” agreement in which the relationship between the controller and the target is extremely unusual: possessor agreement, copying-to-object constructions and long distance agreement. 78 Theories of agreement Agreement involves the representation of displaced information: some aspect of the controller is marked on the target. This information may be categorised into different types, or features: the main features of agreement are gender, number and person. Each feature has different values and these values determine the specific marking that appears in an agreement relation (e.g. first person plural inclusive). Languages vary in both the features that are used and the set of values available for a given feature. The features that enter into a specific agreement relation may depend on specific prerequisites such as the word category, phonological form or lexical information. Additionally, agreement may be subject to syntactic or semantic conditions: considerations such as animacy and topicality may affect the behaviour of agreement. This framework provides descriptive tools that can accommodate the range of agreement phenomena across the world’s languages, without being too deeply entrenched in any specific theory concerning the nature of agreement (or language structure in general). As stated at the beginning of this chapter, this offers a way of describing agreement in a relatively undocumented language, in this case LSE, in such a way that we can compare it to agreement in other languages. A further means of assessing agreement in LSE is provided by the notion of canonicity developed by Corbett (2003b, 2006) on the basis of the descriptive framework already described. Table 2.3 contains the criteria for canonicity, which set out the properties of agreement in its most agreement-like manifestation. As well as describing agreement in LSE and placing it in the context of agreement cross-linguistically, a further object of this study is to examine the role of agreement within the language system, and specifically the extent to which it forms part of the grammar’s syntactic mechanisms. The Minimalist Program has developed a theory of agreement that distils the properties of agreement down to the barest syntactic terms, thereby converting it into one of the fundamental operations carried out by syntax. The next section presents the conception of agreement within the Minimalist Program. 2.3. The Minimalist Program As the latest incarnation of the generativist school of linguistics, the Minimalist Program is the product of a research tradition that focuses on the nature of language as a unique cognitive capacity of humans. The guiding principle behind the generativist approach is the notion that language is a system that can be described in terms of a set of rules: these rules determine what is permissible and should produce correctly-formed sentences in the The Minimalist Program 79 language. Generativism has been applied across different areas of linguistic research, from phonology to language acquisition, but much of the body of work has centred on syntax (and its interaction, or interfaces, with other linguistic levels, such as semantics and phonology). Initially, rules were of the form N VP  S (“put a noun and a verb phrase together and you get a sentence”), but they have since evolved to a much greater level of abstraction expressed in terms of the structural relations between elements (as will be exemplified in the explication of agreement from a minimalist point of view in section 2.3.3 below). Recall that for the study of agreement in LSE, the Minimalist Program is of relevance on two main counts. Firstly, because agreement has taken centre stage within the Minimalist Program and is judged to be one of the basic operations used by syntax to create well-formed sentences. As a result, agreement is defined in very specific (syntactic) terms. Secondly, the Minimalist program stems from a tradition that asks fundamental questions about the characteristics of language as a cognitive capacity. These questions tie in well with those that underlie this study of agreement in LSE concerning the way in which the use of space in a (signed) language is accommodated by the language system. The presentation of the Minimalist Program is organized as follows: in section 2.3.1, I give a brief background to the generativist tradition, providing an overview of the important issues and the developments that have shaped current thinking and that are relevant to the Minimalist Program. Section 2.3.2 presents the architecture of the language faculty as envisaged by minimalism. Finally, the theory of agreement within minimalism is set out in section 2.3.3. 2.3.1. Generativism: issues and developments A central concept that has motivated the Chomskian revolution in linguistics and guided the generativist enterprise is the notion of Universal Grammar (UG), the idea that at its core language is a cognitive capacity with a fixed set of properties. Individual languages may differ in details, but all languages share a common set of properties that are shaped and limited by UG. Furthermore, Chomsky (1965) makes strong claims about the nature of UG and maintains that it is an innate faculty that is hard-wired into the human brain. This conceptualization of language as a limited set of rules leads to the search for those rules. Initially, rules expressed the means by which syntactic elements could be combined and manipulated to form sentences. This gave rise to the development of X-bar theory, which provided the basic framework 80 Theories of agreement for creating syntactic structures, most commonly represented in the form of tree diagrams. (39) a. b. More specifically, X-bar theory provides a greater level of abstraction than merely formulating rules, as it characterizes the way in which rules are The Minimalist Program 81 constrained. The basic unit of X-bar structure is the building block for a recursive configuration that can account for the arrangement of elements in a sentence. (39a) shows a head X projecting a maximum projection of the same type, XP. The complement and specifier positions may be occupied by other maximal projections, thus making it possible for one structure to be nested inside another. Applying this basic structure to the syntactic analysis of an English sentence produces a tree diagram as shown in (39b). The positions within the syntactic structure are occupied by lexical elements such as verbs, nouns, and adjectives, which project verb phrases (VP), noun phrases (NP) or adjectival phrases (AdjP), but there are also positions that are functional in nature and serve to account for the role in syntax of elements such as inflection (IP), and complementizers (CP). IP initially provided the structure required for a finite verb to acquire its inflected form. Subsequently, the split-Infl hypothesis for clausal structure subdivided this part of the structure into various projections such as TP (for tense), AgrSP (for subject agreement) and AgrOP (for object agreement) (Pollock 1989, Kayne 1989, Belletti 2001). Of relevance here, the AgrSP and AgrOP projections were dedicated explicitly to accounting for agreement phenomena (but, as shall be explained in section 2.3.3, they were subsequently done away with). Furthermore, the verb itself has more structure than a simple projection: in addition to VP, a higher vP projection (or “shell”) dominates VP. The lexical verb occupies VP while a light verb, and by extension verbalizing affixes, may occupy vP (Hale & Keyser 1993; Chomsky 1995). In the same manner, CP, also known as the left periphery, has been expanded into a series of specific functional projections, but details will not be given here as they do not bear directly upon the analysis of agreement (for details see Rizzi 1997, 2004). The basic clausal structure with the projections that will be of relevance in this study is shown in (40). Another important development in the repertoire of functional projections that populate the syntactic structure is the introduction of the determiner phrase (DP), which dominates the noun phrase. Furthermore, parallels have been drawn between the internal structure of the determiner domain and the clausal domain (Abney 1987; Ouhalla 1991; Aboh, Corver, Dyakonova & van Koppen 2010), with specific functional projections for number (NumP) and a nominalizing equivalent of v (nP). These projections are described in more detail in sections 7.1.1 and 7.1.2 when looking at the location of the ϕ-features within the DP. The internal structure of DP adopted in this thesis (showing only relevant projections) is displayed in (41). 82 Theories of agreement (40) (41) The Minimalist Program 83 The Minimalist Program aims to make generativist theory as parsimonious as possible, and is driven by considerations of economy. The underlying notion is that language achieves its ends with the fewest possible resources. This has brought about a reconsideration of the theoretical apparatus required to account for syntactic phenomena, and a rejection of unnecessary baggage. (As we shall see when we look at agreement in Minimalism in section 2.3.3, one of the victims of this purge for economy is the set of Agr projections.) One of the guiding principles for Minimalism is the Inclusiveness Condition, which states that no new features are introduced by the computational system (Chomsky 2000: 113). This means that syntax must make do with the set of lexical items that appear in the numeration: it may manipulate the items by means of syntactic operations, but may not add anything else in the process of the derivation. The Inclusiveness Condition calls into question the validity of X-bar theory: the three different levels for each projection (X, X’ and XP) do not figure as part of the lexical entry of a given item and must be added during the derivation, thus contravening the Inclusiveness Condition. Rather than define the differences between syntactic objects in terms of their intrinsic features (as is done in X-bar theory), a relational view of projections obviates the need to add unnecessary labels. Under this perspective, a minimal projection (X) is a lexical item that has been selected, a maximal projection (XP) is a syntactic object that does not project, and an intermediate projection (X’) is a syntactic object that is neither a minimal nor maximal projection (Chomsky 1995). 10 As a result, X-bar theory gives way to bare phrase structure, a more streamlined characterization of the way syntactic elements are configured. (Note that this is a change in the way of conceptualizing the structure and how it is represented by the computational system. It is still common practice to refer to and to represent X-baresque positions, even though the underlying concepts depend on bare phrase structure. This is the practice I adopt here.) Bare phrase structure includes only lexical features and the objects constructed from them. Syntax constructs objects from the basic elements taken from the lexicon by means of syntactic operations: Merge and Move. Merge is more basic, and is a recursive, two-place operation that combines two elements to form a larger one. The properties of the resulting object depend on those of the elements that are merged. Merge is essentially the 10 Additionally, the need for labels themselves has been brought into question. As an alternative, category labels may be replaced by sets of grammatical features which characterize the idiosyncratic properties of individual words. Also, labels may represent a violation of the Inclusiveness Condition (Uriagereka 2000). 84 Theories of agreement simple mechanism by which words (or more specifically items from the lexicon) are put together in a structured way to create sentences. In contrast, Move allows an element to change location during the derivation, as occurs with wh-movement, to give a facile example. Move is not a basic operation and can be decomposed into more fundamental steps. Movement does not simply place an element in a new position thereby leaving the original position empty, but in fact it leaves behind some sort of residue (known as traces in earlier versions of syntactic theory). This is attested by the fact that the residue or trace left by the movement creates effects such as cliticization blocking or may even be partially present in non-standard language, as can be heard in auxiliary copying in child language and preposition copying in speech errors. In (42a) the deleted copy of should (represented as ‘should’) prevents the auxiliary have from cliticizing onto the pronoun they; in (42b), a two-year-old repeats the auxiliary verb can in a question in the position that it would occupy in a declarative sentence; and in (42c) a radio reporter repeats a preposition that has already been moved to the beginning of the relative clause (examples taken from Radford 2004: 157, 156, 192). (42) a. Should they should have called the police? (*Should they’ve called the police?) b. Can its wheels can spin? c. Ikea only actually has ten stores from which to sell from. To account for this, movement is considered to be made up of two operations: Copy and Merge. The element to be moved is copied, and then the copy is merged into the new position. To complete the Move operation, some sort of deletion mechanism is required, otherwise there are two instances of the moved element: the copy in the new position and the original in the initial position. The deletion process may be postulated as a separate operation (Chomsky 1995: 400) or as a failure of the original to be given phonological form. The important point is that the deletion process cannot be absolute since the original material affects other processes – such as cliticization, as in (42a) – but at the same time the item does not appear in its initial position in normal speech – in contrast to (42b, c). An alternative explanation is that the deletion occurs at a specific point in the derivation such that it is deleted after it has had an effect, but before it is assigned phonological form. In this section we have seen that the generativist tradition of linguistics revolves around the search for rules of syntax, or more generally, the way in which those rules are constrained. X-bar theory provided a means for characterizing the rules of syntax, but considerations of parsimony introduced by the Minimalist Program have led to the development of bare phrase The Minimalist Program 91 depends on the language: English verbal agreement involves person and number, whereas in Arabic it also involves gender (Nasu 2001). (48) a. To provide a concrete example of the Agree process in action, let us look at a simple case of subject-verb agreement in Spanish. In the sentence Saioa habla francés (‘Saioa speaks French’), the verb ‘speak’ agrees with the subject ‘Saioa’ in person (third) and number (singular). In the derivation, the different elements are merged into the structure from the numeration as shown in (48a). 16 The verb habla merges with its direct object, the DP francés, and continues to merge with (empty) structure to form a syntactic object (v’). The 16 The internal structure of the DPs is not shown here as the main aim is to illustrate how the agreement mechanism works in terms of the functional heads involved in verbal agreement. The question of the DP-internal structure and location of ϕ-features is addressed in section 7.1.1. 92 Theories of agreement subject DP, Saioa, merges with this object to form a maximum projection, vP. This DP has interpretable inherent features number and person, shown in square brackets. (Note that the object DP also has interpretable inherent features, but these are not shown here in order to focus on the subject agreement process). The vP merges with the minimal projection, T, which hosts uninterpretable, unvalued features that require checking and are marked as u in (48a). Finally, the resulting syntactic object, T’, projects a maximal TP. (48) b. The agreement process occurs in order for the unvalued features on T to be valued and marked for deletion. Acting as a probe, T searches within its domain (vP) for a set of valued features. The nearest possible candidate is the subject DP, Saioa, which has a full set of interpretable features (person and number). (The object DP also has a full set of interpretable features, but the intervening subject DP blocks it from acting as the goal.) This probe-goal The Minimalist Program 93 relation is shown by the dotted line in (48b). The uninterpretable features on the probe are valued and marked for deletion, indicated by the crossed out features in (48b). Subsequently, the verb undergoes head movement to v and from there to T. As a result, the valued features on T are affixed to the verb so that the correct form is produced at PF after Spell-Out. Finally, the subject DP undergoes movement to the Spec-TP position (to fulfil an independent condition, the Extended Projection Principle), yielding the correct word order: Saioa habla francés. This final stage of the derivation is shown in (48c). (48) c. Note that this derivation also works under multiple Spell-Out and cyclic phases with minor adjustments to the order in which operations are applied so as to respect each cycle. Since vP represents a phase, the head movement of 94 Theories of agreement V to v must occur before the first phase completes and sends its complement, VP, to Spell-Out. Once this happens, both the subject DP and the verb are within the phase edge (see (46) above), and thus available to operations in the following phase. This means that the Agree operation can take place since the goal is visible to the probe, as can the remaining movement operations. This revised characterization of agreement in terms of the Agree operation does away with the need for movement to establish an agreement relation. 17 Although there was movement in the Spanish example in (48), the Agree operation did not depend on this and the movement took place due to independent considerations. Agree is essentially an operation that deletes the uninterpretable features on the probe, and is, Chomsky claims, specific to language, unlike Merge, which has analogues in other cognitive domains. In this section we have looked at the Agree operation as formulated by the Minimalist Program. This operation establishes a relationship between an uninterpretable feature located in a core functional head (the probe) and an inherently valued feature (the goal) within a specific syntactic configuration. As a result, the uninterpretable ϕ-features on the probe are deleted. Agree is a basic syntactic operation that is unique to the language faculty. 2.4. Summary In this chapter, I have presented two very different – but not incompatible – views of agreement that have been developed on the basis of data from spoken languages. The first, which I have called the typological approach, aims to capture the diversity of agreement phenomena in the world’s languages based on a very open definition: systematic covariance between the properties of two linguistic elements. The second is the characterization of agreement as a fundamental syntactic operation that is at the core of (and perhaps unique to) the language faculty, as developed within the framework of the Minimalist Program. Despite the disparity between these two approaches, both offer frameworks within which to examine agreement in LSE. In addition, they provide a series of tools that will serve to analyse the phenomenon and to judge the degree to which agreement-like processes in LSE are the same as what is described as agreement in spoken languages. 17 Under the current analysis, Agree becomes a prerequisite for movement since both Merge and Agree are each components of Move: Move establishes agreement between α and F and merges a phrase determined by F to a projection headed by α (Chomsky 2000: 101). The copying component of Move mentioned at the end of section 2.3.1 has been reformulated as an Agree relationship plus some other process that determines the phrase that enters the Merge operation. Summary 95 When we come to examine the LSE data, these aids will bring us closer towards answering one of the basic research questions motivating this study: are we talking about the same thing when we describe agreement in signed and spoken languages? Before turning to the LSE data and evaluating it using the contributions of theories developed by the study of spoken languages, the theoretical background for this thesis would not be complete without looking at the work that has been carried out on other sign languages. Although research into sign language is a much younger field, there is a considerable body of work related to agreement in several sign languages and various theories have been developed concerning verbal agreement. The next chapter presents these theories. 97 3. Agreement in sign languages The previous chapter reviewed the literature on agreement as studied for spoken languages. This chapter overviews the diverse manifestations and analyses of agreement as have been proposed in the literature for different sign languages, and attempts to take in as many agreement-like phenomena from the repertoires of sign languages. The objective, as set out in section 1.5, is to look at all possible candidates for an agreement mechanism in signed languages based on what has been described in the literature, before turning (in chapter 5) to the specific signed language that is the focus of this study, Spanish Sign Language (LSE). This will put us in a position firstly to situate the LSE data in preceding work on other sign languages and secondly to assess to what extent agreement in LSE resembles agreement in other sign languages. This will also lay the groundwork for comparing what is called agreement in LSE with agreement as understood for spoken languages by adopting the frameworks introduced in chapter 2. As mentioned in chapter 1, most research attention has been focused on agreeing verbs and they will also take up much of this chapter. The agreeing mechanism underlying agreeing verbs is spatial, and (as outlined in section 1.3) the verb changes certain aspects of its form in order to indicate one or more of its arguments. The agreement process relies upon an association between a referent and a point or location in the signing space. A more basic use of this association between referent and location is seen in the pronominal system, and in many ways pronominal reference underlies the verbal agreement system. For this reason, in section 3.1 pronominal reference and the nature of the spatial reference system will be described before looking at verbal agreement proper in section 3.2. The description of agreeing verbs includes a detailed look at prototypical agreeing verbs, and the interesting properties they show, as well as backward agreeing verbs, which provide an opportunity to review previous analyses for this type of directional verb. 1 1 This review is not exhaustive as there has been much work on verbal agreement in sign languages, and I limit myself to those analyses that are relevant to this study. One notable absence is Liddell’s (2000, 2003) work that calls into question the linguistic status of this spatial “agreement” mechanism and has undeniably catalysed much work in this field. Nevertheless, I refer to Liddell’s work where relevant and many of the issues raised by 98 Agreement in sign languages Section 3.2 also includes another type of verb that shows spatial agreement not normally included in analyses of verbal agreement in sign languages. This process, which I refer to as single argument agreement, occurs when a verb is articulated at a location in signing space to agree with just one argument. Continuing within the verbal domain, section 3.3 deals with agreement auxiliaries and describes the different types of auxiliary verbs that mark agreement that have been attested in the sign language literature. The spatial agreement mechanism described in this chapter may also be expressed by non-manual features that can indicate locations in the signing space, such as head tilts or eye gaze. Evidence for such non-manual markers of agreement is examined in section 3.4. Just as agreement is not restricted to the verbal domain in spoken languages, this spatial agreement mechanism in sign languages may also be exploited for agreement in other domains; in section 3.5 we consider this possibility for the nominal domain and look for evidence of DP-internal agreement. The chapter closes with a section that summarizes the main characteristics of spatial agreement in sign languages. 3.1. Pronominal reference The pronominal system in most sign languages is most crudely described as pointing. In the case of physically present referents, the signer points towards the referent in order to achieve deictic reference. For non-present referents, the signer’s point is to a location associated with that referent (Cormier 2012). To go beyond a mere crude description of pronominal reference in sign languages, various refinements are necessary. Firstly, pointing may be done in various ways: manually, most commonly with the extended index finger (B), but also with the ] handshape or the 2 handshape (Pfau 2011). In some sign languages (e.g. Libras, Berenz 2002), the handshape used for pointing may change according to style or register. 2 In the non-manual domain, pointing may be carried out by means of a head nod/tilt, eye gaze or even by lip pointing, as described for Providence Island Sign Language (Washabaugh 1986, cited in Meier & Lillo-Martin 2010: 348) and for an idiolect of ASL (Bahan 1996: 86fn). Secondly, in the case of non-present referents, there are various ways in which the referent may be associated with a location in space. The general process of making use of a point in signing space has been called Liddell have stimulated (and are, to some extent, I hope, addressed by) the analysis offered in this thesis. 2 The non-dominant hand may also be introduced to “shield” the pointing hand when using the B handshape in order to make the signalling more discrete. This form has also been reported as a polite pronoun in some sign languages, such as Libras (Berenz 2002). Pronominal reference 99 “indexing” (Friedman 1975) or “nominal establishment” (Klima & Bellugi 1979). The specific process by which a referent is associated with a location in space, which I shall refer to as location assignment, is described in the next section. Finally, the association between a location in space and a referent is discourse-dependent: the association is created and valid for a given stretch of discourse, and may change from one discourse setting to the next. Furthermore, the association may change within a given stretch of discourse in systematic ways through the use of role shift, described in section 3.1.2. 3.1.1. Location assignment From the formal point of view, the association between a referent and a location in space may be established by means of a point or index toward that location (which may be any variety of the manual or non-manual types described above) in combination with a nominal sign. This can be seen in the NGT example in (1), which involves the assignment of the nominal INGEBORG to a location to the signer’s right by means of a point immediately after the nominal. (The sentence also contains a second point to the same location, which serves as an anaphoric reference to the same referent.) NGT (van Gijn 2004: 18. Images reproduced with kind permission from the author.) (1) INGEBORG IXx HAPPY BECAUSE IXx GRADUATE ‘Ingeborgi is happy because shei will graduate.’ An alternative strategy, which obviates the need for any pointing, is to articulate the nominal sign directly at a location (which I shall refer to as localization), thus associating the referent with that location. 3 The availability of localization depends on the phonological properties of the sign in question: body-anchored signs, which are articulated in contact with or near to part of the body, cannot be moved out into the signing space. To overcome the immobility of body-anchored signs, a classifier may be used to localize the 3 For the sake of clarity, it should be noted that localization is not necessarily location assignment. A referent may already be assigned a location, and subsequent productions of the localized sign merely refer to that referent. It is the first articulation of the nominal and the location (via pointing or localization) that achieves the association between the two. The continued use of a localized sign may be due to considerations of referential identity and coherence. It is also possible that pragmatic and discursive functions play a role, and the signer may wish to reassert a location assignment to ensure that her interlocutors are keeping track of the spatial distribution of the referents. 100 Agreement in sign languages referent in the signing space. The use of classifiers is particularly suited to spatial descriptions (see section 1.1.1) but they may also be used for referent differentiation in general discourse. 4 Location assignment, then, may involve pointing, localization and classifiers. As far as the choice of location is concerned, various factors may play a role. If there is a location in the discourse setting that the interlocutors associate with a referent (for example, the desk a colleague normally sits at) then that location may be used (Cormier 2012). Often in such cases, the nonpresent referent is imagined as being present at the location and occupies a life-size space (Liddell 1990). For Danish Sign Language (DTS), EngbergPedersen (1993: 71-74) identifies various conventions that may guide the selection of a location for a referent: the iconic convention conditions the choice of locations so that they reflect the actual spatial relationships between the referents; the convention of semantic affinity places semantically related items at the same location in signing space; and the convention of comparison places referents being contrasted with each other on the left-right lateral axis in front of the signer. Furthermore, locations may be chosen according to metaphorical schemes: for example, Nilsson (2008: 53) describes how in a Swedish Sign Language (SSL) discourse describing an interaction between a doctor and a patient, the doctor may be associated with a location higher than that associated with a patient to reflect the power relation between the two. This sort of convention has also been described for other sign languages (e.g. Bahan & Petitto 1980 for American Sign Language – ASL). At a more formal level, Barberà (2014) describes how in LSC different planes in the signing space are used for location assignment according to the semantic properties of the referents, such as specificity. Nevertheless, when such discourse and pragmatic factors do not play a role, it is generally assumed that the choice of locations is arbitrary. In addition to locations in the signing space, referents may also be associated with locations on the non-dominant hand, specifically the tip of each of the fingers. This strategy is normally used when dealing with two or more referents that form some sort of natural class or group, and is known as a list buoy (Liddell 2003: 223). The location assignment may or may not involve pointing, and may make use of the possibilities for simultaneity afforded by having the non-dominant hand do something different from the dominant hand. Similarly to locations in the signing space, once established 4 In the use of classifiers for personal reference, for Turkish Sign Language (TID) a distinction between a neutral and a honorific classifier has been described (Zeshan 2003b: 64-67). The neutral form uses the B handshape, in contrast to the 2 handshape of the honorific form. Agreeing verbs 107 distinctions will be returned to in the discussion of the ϕ-features present in LSE in chapters 5 and 6. 9 The locations employed to track references in sign languages are not pure logical indices since a clean, constant one-to-one correspondence is not adhered to. However, pronominal points are dominantly indexical in nature, as becomes clear when comparing them to pronominal reference in spoken languages (Cormier, Schembri & Woll 2013). Treating locations in space as R-loci that serve as an explicit manifestation of an abstract index (which is mapped onto a discourse referent) provides a framework that can explain the largely indexical nature of pronominal reference in sign languages. Furthermore, the model also accounts for the unlistability, the discourse determinacy and the potential for shift of these forms. This is the model that will serve as the basis for the analysis developed in this thesis. 3.2. Agreeing verbs Verbal agreement has attracted much attention in the sign language literature, and the existence of verbs that show a spatial means for marking arguments has been identified and described for many sign languages (see Mathur & Rathmann 2010, 2012 for overviews of the phenomenon from descriptive and theoretical perspectives, respectively). A striking feature of verbal agreement in sign languages is that not all the verbs in a sign language show agreement. Although most verbs can be modified for aspect and many sign languages display a rich aspectual system (Pfau, Steinbach & Woll 2012), not all verbs can inflect to mark their arguments. Thus, for example, in Mexican Sign Language (LSM), verbs such as KNOW, UNDERSTAND and LOVE do not move between points in space to mark their arguments (Cruz Aldrete 2009). These are known as plain verbs. Of those verbs that are directional, an important distinction is made between spatial verbs and agreeing verbs (Padden 1983/1988 10 ). Underlying this 9 The glosses in the transcriptions distinguish between locations on the signer’s body (by using a 1 subscript) and locations in the signing space (by using a letter subscript, such as x or y). (See the notation conventions for more details.) Although this gives the appearance of a first/non-first distinction, this is done for clarity given the phonemic salience of contact with the body. Using this notation convention does not commit me to a first/non-first person distinction, and, as shall become clear in the analysis developed for agreement in LSE, this is not the path I take. 10 Padden initially distinguished between spatial verbs and inflecting verbs (1983/1988, 1986), but this was merely a question of nomenclature, and in later work (1990, 1998), she adopts the term “agreement verbs” (following Liddell & Johnson 1989). Other terms that are used in the literature, such as “indicating” (Liddell & Metzger 1998 and subsequent work by Liddell) or 108 Agreement in sign languages distinction is a difference in the use of space: spatial verbs use the signing space topographically to describe spatial relations, whereas agreeing verbs use space categorically to mark person and number of the verbal arguments (Padden 1990). This has several consequences for how each type of verb behaves. Spatial verbs (such as MOVE, PLACE, THROW and STOP in LSM) use the entire signing space to show manner and path of movement such that any slight change in the movement of the sign alters its meaning; agreeing verbs (such as HELP, TEACH, PAY and RESPECT in LSM) restrict movement to the horizontal plane and “vary the position of the beginning and end points of the sign depending on the person agreement, an inflectional category to which spatial verbs are oblivious” (Padden 1983/1988: 47). Examples of each type of verb in LSM are shown in (7). LSM (adapted from Cruz Aldrete 2009: 733, 761, 747) neg (7) a. PEPE IXx IX1 NOT-KNOW ‘Pepe doesn’t know me.’ eye gazex b. GLASS IXx TABLE yMOVEx ‘Take the glass to the table.’ c. PEPE xRESPECTy IXy MARÍA ‘Pepe respects María.’ The plain verb NOT-KNOW (7a) cannot inflect to mark its arguments, even though the referents in the sentence are associated with points in the signing space (‘Pepe’ at locus x, and the first person at the signer’s chest). The verb MOVE (7b) is modified spatially to alter the meaning of the verb: the start point of the sign is the locus associated with the addressee, or more specifically, the place where the addressee is, and the end point is the locus associated with the table. Additionally, the end point is also marked by eye gaze towards that location, as marked in the glosses. The arguments associated with the verb are locative, giving a literal meaning of ‘Take the glass from the place where you are to the table’. In contrast, the arguments of the verb RESPECT (7c) are not locative but person arguments: the verb starts at the locus associated with the subject argument (PEPE) and moves to the locus associated with the object argument (MARÍA). “deictic” (Morales-López et al. 2005) verbs, reflect the framework and analysis adopted by the authors, and will be mentioned where relevant in the description of theories of verbal agreement in sign language throughout this section. Agreeing verbs 109 These agreeing verbs, as they are now most commonly known, are described in section 3.2.1, with an overview of the difficulties in providing a systematic account of the phenomenon and of the different theories that have attempted to characterize this verbal process. Additionally, the existence of certain “backwards” verbs that show the reverse pattern of association between start/end point and subject/object makes finding a coherent account even more challenging, and this is dealt with in section 3.2.2. Although work on verbal agreement is normally restricted to directional verbs with two arguments, some verbs make use of a spatial mechanism to mark just one of their arguments, and this mechanism is described in section 3.2.3. 3.2.1. Prototypical agreeing verbs The spatial behaviour of multi-directional (Friedman 1976) or directional (Fischer & Gough 1978) verbs, as they were first described, has been the focus of most of the work on verbal agreement in sign languages. The phenomenon of interest is shown by the different realizations of the verb TEASE in New Zealand Sign Language (NZSL) in example (8). A change in the direction of the verb changes the arguments that are identified as the subject and object: in (8a), the verb moves from a point associated with the first person subject, to a point associated with a non-first person object, yielding ‘I tease you’; while the inverse movement (and, in this case, orientation of the hand) gives the meaning ‘You tease me.’ NZSL (Online Dictionary of New Zealand Sign Language)11 (8) a. 1TEASEx b. xTEASE1 ‘I tease you.’ ‘You tease me.’ The means these verbs use to mark agreement manually, by movement and/or orientation, are described in section 3.2.1.1., as well as marking for 11 Images taken from videos of sentence examples in the Online Dictionary of New Zealand Sign Language, available at http://nzsl.vuw.ac.nz/. 110 Agreement in sign languages plurality. Section 3.2.1.2 addresses the fact that agreement does not always appear, neither on all the verbs of a given sign language nor all the time on a given agreeing verb, describing agreement marker omission and defective agreement, two phenomena that contribute toward the optionality of agreement. The restriction of agreement to certain verbs is explored in section 3.2.1.3 by looking at the possible prerequisites and conditions that apply to agreement and that could offer an explanation as to why agreement is sporadic in sign languages. Any attempt to restrict agreement to agreeing verbs inevitably leads to assessing what sets them apart from the other group of inflecting verbs, namely spatial verbs. Section 3.2.1.4 re-examines the distinction between agreement and spatial verbs, and refers to analyses that treat all inflecting verbs (both agreeing and spatial) as a single category. A summary is given in section 3.2.1.5. 3.2.1.1. Marking agreement Agreeing verbs in sign languages mark two arguments by moving between loci associated with the verb’s arguments, specifically from the subject locus to the object locus. However, this is not the whole story. The different forms of the NZSL agreeing verb shown in (8) change in the direction of the movement but also in the orientation of the (dominant) hand. RSL (Schwager & Zeshan 2008: 536. Images reproduced with kind permission from John Benjamins Publishing.) (9) a) 1DISTURBx b) xDISTURB1 ‘I disturb you.’ ‘You disturb me.’ For some verbs, orientation alone may be used to mark agreement, typically with the palm or the fingers of the hand facing towards the object locus and away from the subject locus. This is the case of the Russian Sign Language (RSL) verb DISTURB, which includes a downwards movement and so uses only Agreeing verbs 111 orientation to mark agreement, as shown in (9). 12 The distinction between movement and orientation is central to some characterizations of agreeing verbs that will be discussed in section 3.2.2. Whether or not the inflected verb form moves through the signing space, the important idea is that its form aligns with a vector between the two loci associated with its arguments (Padden 1990). As such, agreeing verbs are always transitive or ditransitive verbs, since two arguments are required for agreement to take place. In the case of ditransitive verbs, the second argument that is marked is invariably the notional indirect object rather than the direct object (Cormier, Wechsler & Meier 1999), as shown in (10) for the LSM ditransitive verb GIVE. As mentioned in section 2.2.4, this distinction between primary (direct object in a monotransitive clause or indirect object in a ditransitive clause) and secondary (direct object in a ditransitive clause) objects is not uncommon in spoken languages (Dryer 1986). LSM (adapted from Cruz Aldrete 2009: 749) (10) IXx IXy BOOK xGIVEy ‘She gave him a book.’ The claim for agreeing verbs is that they mark the person and number features of their arguments. The issue of person marking has been mentioned already in the discussion of pronominal reference in section 3.1.3; the manifestation of number will be described here. Number and plurality are marked in a variety of ways by sign languages, and in the verbal domain there are various possibilities for distinguishing between singular and plural referents. Singular has zero marking, a strategy that is extremely common in spoken languages also (Steinbach 2012). For non-singular referents, different forms have been described: multiple or collective marking by means of an arc in the horizontal plane; distributive or exhaustive marking by means of multiple reduplication together with a sideward movement; dual marking by a single reduplication (with a change of location between each realization), or by simultaneous or sequential realization on each hand (Padden 1983/1988; Mathur & Rathmann 2010: 181). The multiple and distributive forms for the Australian Sign Language (Auslan) verb ASK are contrasted with the singular (zero marked) form in (11). The distributive form (11c) is very similar in form to certain types of aspectual marking, such as the iterative (Wilbur 2008). Additionally, the modification relates to the temporal structure and properties of the event 12 For more examples with photos of different agreeing verbs that contrast the use of movement and orientation, see (8) in section 5.2.1. 112 Agreement in sign languages being described, as proposed by Wilbur’s (2008) Event Visibility Hypothesis. Taken together, these observations suggest that the distributive marking is as much about the event expressed by the verb as about its arguments. Recall that in section 2.2.5.2 a distinction was drawn between nominal number – relating to the arguments controlling the agreement process – and verbal number, which is an inherent feature of the verb. The distributed form, insofar as it involves reduplication of the verb, appears to provide information about verbal rather than nominal number. For the time being, I focus on the multiple form (11b) as the marker on the verb of plurality of the verb’s argument, but will return to the issue of these different plural markers in section 6.4.2 in the assessment of the expression of number in verb agreement in LSE. Auslan (Johnston & Schembri 2007: 148. Images reproduced with kind permission from Cambridge University Press.) (11) a) 1ASKx b) 1ASKx.PL c) 1ASKx.DISTR ‘I ask him.’ ‘I ask them.’ ‘I ask each of them.’ 3.2.1.2. Lack of agreement marking Verbal agreement in sign languages shows several interesting properties related to the optionality of the appearance of agreement. Firstly, not all verbs in a given sign language show agreement. Secondly, those verbs that are agreeing verbs often do not show agreement for one or even both of the arguments. Finally, some agreeing verbs show a defective paradigm due to the fact that their phonological form obstructs the appearance of the marking for the subject argument. Each of these phenomena will be described in turn. As pointed out in the introduction to this section, agreeing verbs constitute a subset of the verbs, which means that this mechanism is not uniform across all verbs in a given sign language. Cross-linguistically, this is extremely unusual since if a language has verbal agreement, it appears on all verbs (Corbett 2003): although there are two spoken languages that show Agreeing verbs 113 agreement on only 30% of the verbs, the Nakh-Daghestanian languages Chechen and Ingush (Bickel & Nichols 2007), this is a very rare situation. Within the class of agreeing verbs, there is a notable tendency not to show agreement. For ASL, Meier (1982) and Padden (1983/1988) noted that the marking of both arguments is not obligatory and the verb may inflect for just one of its arguments. If this happens, the argument that is omitted is always the subject, as shown in (12). The agreeing verb GIVE starts not at a locus associated with the subject argument WOMAN but at a neutral location in the signing space. This phenomenon of agreement marker omission is also attested in other sign languages: example (13) from LSM shows how an agreeing verb may mark only the object argument (compare with (7c) above, in which the verb RESPECT marks both arguments). This will prove important for characterizing agreeing verbs when considering backwards verbs, described in section 3.2.2. ASL (adapted from Padden 1983/1988: 136) (12) WOMAN neutGIVE1 NEWSPAPER ‘The woman gave me a newspaper.’ LSM (adapted from Cruz Aldrete 2009: 739) (13) JUAN RESPECTx ‘Juan respects you.’ Additionally, agreement may be completely absent and neither argument is marked on the verb. Corpus-based studies in various sign languages have revealed that that agreeing verbs more often appear uninflected than inflected, even in contexts in which agreement marking could be expected (de Beuzeville, Johnston & Schembri 2009 for Auslan; Schuit 2013 for Inuit SL). The optionality for marking one of the arguments, described above as agreement marker omission, should be distinguished from the case of transitive verbs that can only show marking for one argument, which has been called single agreement (Meier 1982). In order to avoid confusion with the distinct phenomenon of single argument agreement, to be introduced in section 3.2.3, I use the term defective agreement to refer to this (phonological) restriction of agreement to a single argument. In the case of agreement marker omission, marking for one of the arguments may not appear; for defective agreement, marking for one of the arguments cannot appear. This occurs with verbs such as SEE or TELL in ASL (Hahm 2006), which cannot show marking since the form of the verb includes a location (near the eyes and near the mouth, respectively, for these verbs). As a result, these verbs show a defective agreement pattern in which only the object is marked. In this sense, the 114 Agreement in sign languages phonological form of the verb is a prerequisite for agreement: the verb must have no specifications for location in order for full (i.e. two-place) marking to take place. This issue of defective agreement will be returned to in the next section when we look at the phonological constraints that operate on agreeing verbs. 3.2.1.3. Prerequisites and conditions As the previous section showed, agreement in sign languages is both sporadic, in the sense that only some of the verbs show agreement, and optional, since agreement may or may not appear. Leaving aside the issue of the optionality of agreement, which will be addressed in chapter 6, the uneven distribution of agreement across the verbs of sign languages may be due to various constraints on agreement in the form of prerequisites and conditions. In section 2.2.6, we saw that prerequisites operate on agreement at the phonological and morphological level, while conditions are considerations of a syntactic or semantic order that have an effect on agreement. Considering possible conditions on agreement, a syntactic condition could be that agreeing verbs must be transitive or ditransitive since two arguments must be available for agreement to take place. This condition is implicitly included in many authors’ definitions of agreeing verbs, whether they mention agreement with subject and object (Padden 1983/1988: 47; Meir 2002: 421) or make explicit mention of two arguments (Edge & Herrmann 1977: 147; Mathur 2000: 212; Hong 2008: 170). As we saw in section 3.2.1.2, defective agreement occurs when arguments cannot be expressed on the verb (for phonological reasons) but the verb is still required to have two (or more) arguments. Limiting agreement to verbs with two or more arguments is a somewhat arbitrary state of affairs, and as I shall suggest in section 3.2.3, agreement can and does occur on verbs with a single argument. Furthermore, the condition would be necessary but not sufficient for agreement to occur: many plain verbs are transitive, such as KNOW in (7a) above. A semantic condition imposed on agreeing verbs is the requirement that both the verb’s arguments be [+human], or less restrictively [+animate]. Some authors include such a condition in their definition of agreeing verbs (e.g. Mathur 2000: 212; Hong 2008: 170). 13 This semantic condition may be subject to cross-linguistic variation since there are clear differences in the literature. 13 This restriction seems to be related to the notion of transfer mentioned below in section 3.2.1.4: arguments which can be the source or goal of transfer are either locative (in which case the verb is spatial) or potential possessors (and therefore most likely [+human], or at least [+animate]). This idea is picked up in Meir’s (1998b) analysis, which will be described in section 3.2.2. Agreeing verbs 115 For example, Yang & Fischer (2002: 171) report an agreeing verb in Chinese Sign Language (CSL) that can inflect for the object only if the object is animate. In contrast, for LSQ, we find an example of an agreeing verb marking for two inanimate arguments: in (14) the verb FEND-OFF marks two non-first person arguments, which are ‘the ozone’ (presumably mentioned earlier in the discourse) and ‘the solar system’. LSQ (adapted from Nadeau & Desouvrey 1994: 153)14 (14) xFEND-OFFy SOLAR-SYSTEM ‘[Ozone] protects us from the solar system.’ Alternatively, one way to explain the distribution of verbal agreement might be that a phonological prerequisite excludes many verbs from agreeing. As we saw above in the description of defective agreement (section 3.2.1.2), verbs with a specified location cannot show full agreement. Thus, since agreement may be manifested through movement and/or orientation, it may be that verbs for which both movement and orientation are specified in their phonological form cannot show agreement. The full specification of location (i.e. for all the location slots of the sign’s phonological matrix) may also bar agreement since this effectively limits movement and makes it unavailable for the expression of agreement. This would go a long way to explaining why body-anchored verbs are plain verbs. However, it would do little to explain why spatial verbs do not show agreement. The distinction between spatial and agreeing verbs is questioned in the next section. In trying to distinguish agreeing verbs from other verb classes we have looked at (syntactic/semantic) conditions and (phonological) prerequisites that might determine the appearance of agreement. We now focus on the class of agreeing verbs themselves to consider phonological constraints to which they are subject. These verbs fall into two categories: the first, already mentioned above, is defective agreement, which occurs when an agreeing verb cannot inflect for both arguments; the second arises due to gaps in the agreement paradigm caused by phonologically illicit forms for certain person/number combinations of the arguments. In order for agreement to take place, a verb must be able to move through space (or at least to orient itself in space). However, many verbs have 14 In the original example, Nadeau & Desouvrey give the glosses in French, and the verb is glossed as DEFENDRE. I gloss the verb as FEND-OFF in order to show the argument structure more clearly, since the subindices on the verb indicate that it agrees with two third person arguments and not with ‘us’, which the authors include in the English translation of the sentence. The important point is that the verb is agreeing with inanimate arguments. 116 Agreement in sign languages a lexically specified location that constrains the start point of the sign, thus blocking the possibility for marking agreement with the subject argument. This gives rise to a defective agreement paradigm. In her analysis of the phonological clashes that arise between verb roots and the movement needed for agreement, Meir (1998b: 90) mentions that several such verbs exist in Israeli Sign Language (ISL), such as ASK and ANSWER (both specified near the mouth), SEE (near the eye) and TELEPHONE (near the ear). These verbs generally agree with only the object argument, and the subject argument must be marked in some other way (by the presence of an explicit nominal or pronominal), as shown in (15) for the defective agreeing verb SEE in ASL. ASL (adapted from Cormier, Wechsler & Meier 1998: 220) (15) IXx SEEy ‘She sees him.’ In ISL, though, certain forms of these verbs may show agreement with both arguments: [t]he only forms of these [defective] verbs which agree with two arguments are those that inflect for 1P object. In such cases, the verb form has a complex path movement: it begins [at the locus associated with] the subject, moves to the specified location, and then to the 1P locus (the signer’s chest). (Meir 1998b: 90-91) When describing defective agreeing verbs in ASL, Mathur & Rathmann (2010: 178) mention that a variant of the verb TELL does allow agreement with both arguments: in contrast to ISL, which includes the specified location as the intermediate point of the sign (subject>mouth>object), the strategy in ASL is to begin the sign at the lexically specified location and then move to subject locus and from there to object locus (mouth>subject>object). The second type of constraint on the form of agreeing verbs arises as a result of articulatory incompatibilities. For example, the ASL sign GIVE is lexically specified as having the palm of the hand orienting upwards, which requires radio-ulnar supination. At the same time, it expresses agreement through movement and orientation. A first person object form would require pointing and moving the hand inwards (towards the signer) and a plural form would involve adding an arc movement. If the target form is first person plural object, this would require both of these movements. However, combined with radio-ulnar supination (in order to keep the palm of the hand facing upward), the result puts the arm in an awkward configuration, and therefore the resulting form is not possible. In order to study which verbs allow which combinations, Mathur & Rathmann (2001, 2006) collected data Agreeing verbs 123 This separate mechanism of facing allows a distinction to be drawn between semantic roles (reflected by the direction of the movement) and syntactic roles (reflected by the facing of the hands). Since Meir’s proposal will be relevant to properties of LSE verbal inflection as analysed in chapter 7, some further details of the proposal will be given here. ISL (Meir 1998b: 84, 123. Images reproduced with kind permission from the author.) (18) a) 1HATEx b) xHATE1 ‘I hate you.’ ‘You hate me.’ c) xTAKE-ADVANTAGE-OF1 d) 1TAKE-ADVANTAGE-OFx ‘I take advantage of you.’ ‘You take advantage of me.’ Based on the observation that agreeing verbs have a specific lexical structure that denotes transfer from a SOURCE argument to a GOAL argument, Meir developed a thematic structure agreement analysis according to which agreeing verbs merge with a particular predicate, DIR (called PATH in Meir 1998b), which denotes movement from one place to another (also present in spatial verbs, which also include the semantic concept of motion). It is this DIR predicate that shows agreement and not the verb itself. To explain the mechanisms of the agreement process, Meir proposes that DIR is a bound morpheme that fuses with the root of the verb, and describes this in terms of a merger of the lexical conceptual structures of each element which results in a complex verb. Essentially, DIR fills the argument slots of the verb root with its own arguments (which have already been assigned thematic roles of SOURCE and GOAL); the complex verb then checks the referential features of its arguments in the syntax (Meir 2002: 438). The referential features of the 124 Agreement in sign languages arguments are expressed as an explicit index in the form of spatial loci and these locations are copied onto the verb’s phonological features at Spell-Out. From a phonological point of view, the DIR morpheme has two empty location slots and may be represented in terms of Sandler’s (1989) Hand Tier Model as in (19): (19) DIR: (adapted from Meir 1998b: 167) μ location [ ] location [ ] After the merger with the verb root, these empty slots serve as the landing site for the phonological location features copied from the arguments’ loci. For this to happen, the verb root must also be underspecified in its phonological matrix for location and path movement, 18 so that the fusion of the verbal root and DIR does not result in a phonological clash. Otherwise agreement cannot be expressed, or is only partially expressed. We have already mentioned cases of this sort: the examples in section 3.2.1.1 showed how agreement by movement may be completely blocked by lexically specified movement (and/or location/contact) in the verb’s phonological matrix, resulting in either a plain verb (such as LOVE in LSM) or a verb marking agreement through orientation alone (such as DISTURB in RSL). Alternatively, movement may be partially blocked (as described in section 3.2.1.2) in the case of defective agreement: a lexically specified location prevents the first slot in the DIR morpheme from serving as the landing site for the location of the SOURCE argument. Meir (1998b, 2002) characterizes the independent mechanism of facing, which marks the object argument, as a case marker rather than an agreement process. The main motivation for this is the fact that in sign languages object marking is more prominent than subject marking, as demonstrated by argument marker omission and defective agreement (section 3.2.1.2). While it is typologically unusual to find agreement with only the object, case marking for only the object is common, so the fact that sign languages mark the object in this way fits into common cross-linguistic patterns of case marking for the object argument. What is slightly unusual is that the case marking appears on 18 Brentari (1998: 4) distinguishes between local and path movement. The distinction is both articulatory and phonological: local movements are made by the wrist, knuckles or finger joints, and may cause a change in handshape or orientation; path movements are made with the elbow or shoulder and may be specified as a movement feature or a change in location. Agreeing verbs 125 the verb rather than on the noun argument itself, but case relations may be marked by verbal affixes in head-marking languages, that is, languages that mark relations on the head (rather than the dependents) of the phrase (Nichols 1986). The case marking by means of facing appears on those verbs whose lexical conceptual structure includes the notion of transfer, thus making a distinction between spatial and agreeing verbs. The former denote motion but not transfer and thus have no facing, whereas the latter do denote transfer, according to Meir’s analysis. Again, if the phonological specification of the verb root contains a specification for the palm and finger orientation, this blocks the possibility of marking the object argument via facing. For ISL, Meir (1998b: 245-52) shows how different phonological clashes restrict the way in which a verb expresses agreement and case marking. One problem with a semantically based account is that it fails to describe the agreement phenomenon in configurational terms; recall from chapter 2 that agreement relationships hold in the context of a specific syntactic configuration. In order to get round this, Meir suggests that agreement is a property of the spatial DIR predicate (present in both spatial and agreeing verbs) and that it is the close relationship between the thematic roles that DIR assigns to its arguments and the agreement slots these arguments are associated with that gives agreeing verbs their thematic flavour. As Meir (2002: 440) puts it, “agreement in ISL is related to specific spatial thematic roles because of the spatial nature of the agreeing element”. Essentially, agreement looks spatial because what agrees in sign languages is a spatial element (DIR). This proposal also deals with another anomaly: as mentioned in section 3.2.1.2, a remarkable property of sign language verbal agreement is that it is not manifested by all verbs in a language, in contrast to the case for spoken languages that whenever a language has verbal agreement, all the verbs show agreement. A consequence of Meir’s analysis is that sign language agreement behaves more consistently since it is no longer a property of a restricted class of verbs (or of any verbs for that matter) but rather it arises whenever a particular predicate (DIR) is present. Another outcome of this model of agreement is a refinement of the semantic restraints placed on the arguments of agreeing verbs. As mentioned in section 3.2.1.3, in the literature it is often claimed that agreeing verbs can only take [+animate] arguments (cf. Janis 1995; Mathur 2000; Rathmann & Mathur 2005; Quadros & Quer 2008); in contrast, following from the characterization in terms of transfer, Meir suggests that the relevant feature is that arguments be potential possessors (1998b: 203 fn). 126 Agreement in sign languages 3.2.2.4. Problems with Meir’s account Although Meir’s analysis distinguishes the movement and facing as two separate mechanisms, treating facing as a case marker effectively excludes it from the agreement process, and the movement element is the only agreement marker. This has two consequences for the analysis. Firstly, the agreement process is common to both spatial and agreeing verbs, despite the differences in the use of space that each type of verb seems to involve (as described in section 3.2.1). Thus, agreement is characterized by the use of loci as the means of exponence whether the verb be spatial or agreeing, in the context of movement between two loci. As we saw in section 3.2.1.4, the distinction between spatial and agreeing verbs is difficult to maintain, and many proposals have grouped together both types of verbs as inflecting verbs (cf. Quadros 1999), so this does not seem to be a drawback. The second consequence is problematic: since movement is closely tied to the (semantic) SOURCE-GOAL relationship (whether this be in terms of motion or transfer), this makes Meir’s account of agreement essentially semantic in nature. The formal device of restricting agreement to a predicate (DIR) with arguments that happen to have specific semantic roles (i.e. SOURCE and GOAL) explains why a syntactic process (agreement) is semantic in appearance and provides a syntactic framework for agreement, but also means that agreement is limited to the semantic context of SOURCE and GOAL. For spatial verbs, this is no issue since a verb that describes movement is necessarily from SOURCE to GOAL. 19 However, Quadros & Quer (2008) point out that (in LSC and Libras) the second argument of an agreeing verb is not always a GOAL, but may be a THEME, as in verbs like PRESS or INVITE. Furthermore, various agreeing verbs, such as CHOOSE or SUMMON, do not have a clear sense of transfer. Meir’s account does not address the question of the features involved in the agreement process and limits itself to stating that the referential features of the language (R-loci) appear as the agreement markers. This leaves the issue of what referential features sign languages use to a theory of reference. However, since the features involved in agreement are central to the process, and as we have seen (in section 3.1), reference in sign language shows very particular characteristics, this issue will be examined carefully in chapters 6 19 A possible exception is the use of movement to trace the shape of a referent, such as a bend in a road, such that the movement describes a path and not motion between a SOURCE and a GOAL. This is normally achieved with classifier constructions (see section 1.2), which use a strongly isomorphic mapping of the signing space onto real space. These structures are usually considered distinct from spatial verbs and the issue is not dealt with in this thesis, but the question of how the use of space in classifier structures interacts with that of lexical verbs deserves further attention. Agreeing verbs 127 and 7, when we assess these agreement-like phenomena in terms of the theoretical frameworks developed for spoken language data. 3.2.2.5. Summary Backwards verbs show the unusual property of inverting the association between the start/end point of the sign and the subject/object argument. Similar lexical items show up as backwards verbs in many different sign languages, though there is variation from one language to another. These verbs make clear that there are two mechanisms at play in agreeing verbs: on the one hand, the movement between the loci associated with the verb’s arguments; and on the other, the facing of the hands towards the locus associated with the object. Backwards verbs are a subset of the agreeing verbs in that they mark agreement for two arguments. The next section looks at verbs that modify spatially to agree with just one argument. 3.2.3. Single argument agreement The verbal agreement mechanism we have examined so far has involved movement (or orientation) between two points in space, and the verb agrees with two arguments. In section 3.2.1.2 we saw cases of verbs marking just one argument (in the context of agreement marker omission or defective verbs) but these were situations in which marking of a possible second argument was omitted or blocked; the verbs are directional but for some reason one of the arguments is not marked. In contrast, in single argument agreement the verb is not directional but localizable: the verb is articulated at the locus associated with the argument. As such, the spatial mechanism employed by the verb only ever allows for one argument to be marked, and only a single agreement slot exists. Example (20) shows how the NGT verb WAIT can be articulated at a point in the signing space associated with a referent, thus showing single argument agreement. NGT (Zwitserlood & van Gijn 2006: 198. Images reproduced with kind permission from Oxford University Press.) (20) a) WAIT b) WAITx citation form ‘He waits.’ 128 Agreement in sign languages The observation that not all plain verbs are body-anchored, and can be articulated at different points in the signing space has appeared frequently in the literature: for example, Fischer & Gough (1978: 22) mention “incorporation of location” in verbs in ASL, and Bergman (1980) uses the term “localization” for verbs in SSL. Bergman’s term for this mechanism fits with the definition of localization already adopted (in section 3.1.1): a sign is articulated at a specific point in the signing space. I use the term to describe the modification of a location of a sign; it does not refer to the function of that modification (see fn. 3 for further clarification). Some authors (Bergman 1980, 1990; Smith 1990; Zwitserlood & van Gijn 2006) consider this phenomenon to be part of the verbal agreement system, on a par with two-place agreeing verbs. However, much of the literature on verbal agreement in sign languages avoids this phenomenon and restricts the discussion of agreement to prototypical and backwards agreeing verbs as described in the previous two sections. For example, a recent landmark paper on agreement in ASL makes passing reference to the issue: Although this class of [plain] verbs is considered non-agreeing, some of them can actually be signed in a locus associated with a location of an event (e.g. WANT, BUY, and LEAVE-ALONE). (Lillo-Martin & Meier 2011: 106) 20 This has led to a certain amount of confusion in the field and deserves attention to make the issues involved explicit. The exclusion of verbs articulated at a single location from the analysis of agreement stems from the observation by Padden (1990) that for transitive verbs such marking is ambiguous since it could be for the subject or the object. This is illustrated by (21), in which the ASL verb WANT is localized at different points in the signing space. The loci may be associated with the verb’s subject arguments or its object arguments, and the interpretation depends on the context. Since the marking of the argument is not systematic, Padden claims that this cannot be a case of agreement. ASL (Padden 1990: 121) (21) WOMAN WANTx WANTy WANTz ‘The womeni,j,k are each wanting.’ ‘The woman wants thisi, and thisj, and thisk.’ 20 Later in the same article, Lillo-Martin & Meier state that “no verbs mark agreement with only the subject (indeed, intransitive verbs are not directional)” (2011: 126) thus identifying agreement with directional verbs. Agreeing verbs 129 However, following up observations made by Engberg-Pedersen (1993), Meir (1998b: 95) points out that the first reading of (21), in which the subject argument is marked by agreement, occurs in a specific discourse context, namely, when several referents in the discourse are being compared. Such contexts give rise to what Engberg-Pedersen calls pragmatic agreement, in which overriding pragmatic considerations indicate which argument is being identified by articulating the verb at a given locus. Outside this specific discourse context, Meir shows that (in ISL) single argument agreement marks the internal argument of the verb, i.e. the subject of intransitive verbs and the object of transitives, as shown in the examples in (22). ISL (Meir 1998b: 94) (22) a. STICK IXx CL(break)x ‘The stick broke.’ b. BOY IXx GROW-UPx ‘The boy grew up.’ c. STICK IXx IX1 CL(break)x ‘I broke the stick.’ d. POLICEMAN IXx THIEF IXy CATCHy ‘The policeman caught the thief.’ This refinement of the characterization of single argument agreement (by excluding the ambiguous cases due to specific discourse considerations) allows Meir to identify a mechanism that uses space to systematically mark arguments at the clausal rather than the discourse level. 21 As such, these verbs that can be localized to mark an argument (and which clearly do not involve the DIR predicate postulated by Meir for agreeing verbs) will be considered as a possible manifestation of (spatial) agreement when looking at the LSE data in chapter 5 and when assessing agreement in chapters 6 and 7. 3.2.4. Summary This section has looked at verbs in sign language that have been described as showing agreement by means of spatial inflection. Most attention in the literature has been given to directional agreeing verbs, which mark for two arguments by moving from the locus associated with one argument to that associated with another. A small subset of these verbs, backwards verbs, 21 Unfortunately, Meir does not analyse this construction in her work on agreement and expressly focuses on directional verbs. 130 Agreement in sign languages shows the property of moving from the object to the subject locus, contrary to the prototypical movement from subject to object locus displayed by most agreeing verbs. These agreeing verbs have been the focus of a great deal of research since they display a series of unusual properties, many related to the distribution of agreement in sign languages: not all verbs can show such agreement, and those that can may omit agreement or have defective inflectional paradigms. Additionally, agreement has a very strong semantic flavour, since it commonly depicts transfer from a SOURCE to a GOAL. This sits uneasily with a characterization of agreement as occurring in a specific configurational or syntactic context. These facts lead Meir (1998b, 2002) to an analysis of agreeing verbs in ISL in terms of a specific spatial predicate, DIR, which shows agreement and fuses with semantically appropriate verbs (those that express transfer). As we have seen, this provides a syntactic framework for a semantically driven process, but may commit agreement to a limited semantic context that data from other sign languages suggest is too restrictive. Finally, we have also looked at the phenomenon of single argument agreement, in which a verb is localized to mark just one of its arguments. This mechanism has been generally overlooked in the literature, but appears to show a systematic use of space to mark a verb’s argument, in the same sense that (prototypical and backwards) agreeing verbs do. As such, it will be included in the possible list of candidates for agreement to be assessed in LSE. The next section continues to look at verbal agreement, but in the context of verbal auxiliaries. The different auxiliaries that have been identified for different sign languages function principally to bear markers of agreement, and so are highly relevant to the issue under discussion. Furthermore, the interaction between lexical and auxiliary verbs provides important insight into the nature of the spatial agreement process in sign languages. 3.3. Agreement auxiliaries In the previous chapter, section 2.2.3.1, we saw that auxiliary verbs are common targets for agreement. In spoken languages, auxiliary verbs generally serve to show information relating to tense, aspect, modality, negative polarity and voice, and the appearance of agreement on these elements is more of a syntactic “accident” (Steinbach & Pfau 2007). In sign languages, however, various elements have been identified that serve as a verbal auxiliary but with the main function of marking subject/object agreement when the lexical verb is not capable of doing so (i.e. when it is a plain verb). Consequently, these elements are referred to as agreement Agreement auxiliaries 131 auxiliaries or by the more specific term “subject object agreement” (SOA) auxiliaries, coined by Steinbach & Pfau (2007: 308). These elements occur together with a lexical verb (to form a monoclausal verb phrase) and perform the grammatical function of marking agreement, and as such they fall under the definition of a verbal auxiliary proposed by Anderson (2006: 5; see section 2.2.3.1 for details). Agreement auxiliaries have been identified for various sign languages, and they can be categorized into three different types based on the interrelated criteria of form and origin. The first type, dealt with in section 3.3.1, normally glossed as AUX, is the most frequent cross-linguistically (based on current data) and consists of an indexical element that appears to be derived from concatenated pronominal forms. The second group of auxiliaries, described in section 3.3.2, is derived from full lexical verbs that have undergone semantic bleaching and taken on a more functional role within the clause. The third type of auxiliary (section 3.3.3), PAM, is similar in function to AUX but its use tends to be more restricted and it appears to have its origins in the nominal PERSON. This section provides an overview and description of each of these types of agreement auxiliary in turn. (For a more detailed overview of auxiliaries in sign languages see Sapountzaki (2012), and for a discussion of the sources from which they grammaticalize, see Steinbach & Pfau (2007)). 3.3.1. AUX Many sign languages have an auxiliary element to mark verbal agreement that consists of signalling the location associated with the subject followed by the location associated with the object. In form, the hand adopts the B shape typically used for pointing and the auxiliary looks like two consecutive points joined by some sort of arced movement. This AUX form was first described for TSL by Smith (1990), shown in figure 3.1, and has subsequently been identified for Argentine Sign Language (LSA) (Massone 1994; Massone & Curiel 2004), Japanese Sign Language (NS) (Fischer 1996), Libras (Quadros 1999; Quadros & Quer 2008), Indo-Pakistani Sign Language (IPSL) (Zeshan 2000a, 2003a), LSC (Quer et al. 2005) and Greek Sign Language (GSL) (Sapountzaki 2005). 132 Agreement in sign languages 1AUXx Figure 3.1 The indexical agreement auxiliary described for several sign languages, here showing movement from first person to a non-first person locus in TSL. (Image reproduced from Smith 1990: 217, with kind permission from University of Chicago Press.) The details of the behaviour of AUX vary from language to language, but basically the auxiliary appears next to the lexical verb and marks the agreement for that verb, as shown in examples (23a) and (24). AUX frequently occurs with plain verbs and serves to express (spatial) agreement that the verb itself cannot inflect to show due to its phonological limitations. In some sign languages, such as GSL and NS, AUX can accompany only plain verbs or agreeing verbs that are uninflected for agreement, and double agreement (manifested on both the main verb and the auxiliary) is not possible, as exemplified by the ungrammatical NS sentence (23b), in which AUX appears together with the inflected agreeing verb HIT. NS (adapted from Fischer 1996: 107) (23) a. CHILDx TEACHERy LIKE xAUXy ‘The child likes the teacher.’ b. *MOTHER FATHER xHITy xAUXy IPSL (adapted from Zeshan 2003a: 172) q (24) UNDERSTAND xAUX1 ‘Do you understand me?’ Targets 235 phrases os garfos vermelhos (‘the red forks’) or as facas vermelhas (‘the red knives’) is very similar in form to the definite articles os/as but this does not mean that the marking involves incorporation of the article into the adjective; rather, both forms make use of same underlying agreement mechanism. This boils down to a distinction between the exponence and the target of agreement, but also requires an analysis of space as the referent marking system at work in sign languages. Essentially, there are two issues at stake here. Firstly, whether or not the marking is some sort of pronoun, and secondly, what the nature of the marking is, an affix or a clitic. The foregoing discussion has shown that that spatial marking behaves quite differently to a pronominal form. A further observation is germane to both issues: location in itself does not constitute independent phonological material. As mentioned above, a location is not a pronoun, but rather a part of a pronoun. By the same token, a location cannot cliticize or form an affix with a verb as it does not have enough phonological material to exist independently. LSE (Ix_lion 1:11) (4) D hand transition(IX1) IX1 1HELPX ND hand transition(1HELPX) 1HELPx -------------------------------------- ‘I’ll help you.’ Cliticized pronouns have certainly been observed in sign languages and two different mechanisms that result in cliticized pronouns have been described (Sandler & Lillo-Martin 2006). In the case of assimilation, the pronoun adopts the handshape of the neighbouring verb; coalescence, on the other hand, involves the production of the pronoun before or after a two-handed sign. The LSE data present both types of cliticized pronoun. Example (4) shows a coalesced pronoun, visible on the dominant hand during the prior transition and the first hold of the bimanual verb HELP as indicated by the handshape (with the index finger selected), demonstrating that the pronoun is clearly 236 Typological perspective of LSE agreement reduced but still present. In contrast, when verbs are spatially marked, no pronoun (or remains of a pronoun) is visible. More recent work has continued to argue that the spatial agreement marking in sign languages behaves more like a pronominal clitic (Keller 2001), drawing on evidence from its syntagmatic properties, distribution, optionality, inventory and realization (Nevins 2011). On the other hand, the evidence presented above – both Corbett’s diagnostics and the argument that the marking does not constitute enough independent phonological material – suggests that the marking is not derived from a pronominal form. While spoken languages commonly use pronouns as agreement affixes, the gestural modality makes available the use of space for referencing and this is recruited for the agreement. In both cases, a referential mechanism becomes part of the reference-marking system on verbs. However, while spoken languages tend to do this with affixes, the form of marking in sign languages is different. The means of exponence for spatial marking in LSE, that is to say, the incorporation of a location or locations in a sign, appears not to involve the addition of phonological material, as is the case with affixes, but rather a stem alternation (Mathur 2000). Phonological models proposed for sign languages include location slots that can be assigned a given value. The spatial agreement process, then, involves assigning a specific value to one of the location slots in the phonological matrix (one in the case of single argument agreement; two for agreeing verbs). However, it should be pointed out that the distinctions in location within the signing space (where location assignment occurs) are not normally relevant as far as the phonology of lexical items is concerned. Most phonologically contrastive locations occur on (or near to) the body. As such, the agreement system makes use of distinctions in form that are not relevant for the phonological system. On a speculative note, it is possible that locations that are not anchored to some point on the body are not specific enough (and therefore perhaps not stable enough) for the phonological contrasts needed by the lexicon. As such, sign language phonologies make use of location but in a restricted domain. However, the spatial medium offers a far greater number of locations. Locations in the signing space can be exploited isomorphically for spatial descriptions, but also offer the possibility of creating a reference tracking mechanism. One can imagine that motivated spatial locations gave way to abstract indexing, by means of a grammaticalization process involving the semantic bleaching of the locative meaning of points in signing space or some basic locative verb BEAT (Wilbur 1999). Such ideas hark back to seminal work by Shepard-Kegl (1985) on space and locatives in ASL, but we have no historical data from LSE to support this speculation. However, evidence from an evolving language Targets 237 system, Nicaraguan Sign Language, suggests that, at least in the case of manual pointing, locative uses are present at early stages of the language and only later do more abstract, nominal uses of pointing appear (Senghas & Coppola 2011). Furthermore, this use of space for both locative and nominal reference ties in well with the fact that the distinction between spatial and agreeing verbs is often blurred, as mentioned in section 3.2.1.4. It also helps to explain why a referential locus can regain its spatial meaning when that information becomes relevant. This is demonstrated by Liddell’s (2000) notorious examples of the type ‘I asked a (tall) man’, in which the sign ASK moves upwards in the signing space to indicate the (relative) height of the person referred to (for a formal semantic characterization of this role of spatial iconicity in sign languages see Schlenker 2011). I return to examples of this type and to the distinction between spatial and agreeing verbs in the discussion of conditions in section 6.5. A unique property of the means of exponence of spatial marking on sign language verbs is related to a property mentioned above in the section on controllers (6.1). The form made manifest on the target is not some feature of the controller per se, such as gender, or even part of the form of the controller, as occurs with radical alliterative agreement (described in section 2.2.3.3). Rather, the target displays a stem alternation based on a form that has been assigned to the controller. In some senses, this is more similar to the type of alliterative agreement seen in Bantu languages, in which a gender marker on the controller appears on the target of the agreement process, as shown in example (5). Swahili (Welmers 1973: 171, cited in Corbett 2006: 87) (5) ki-kapu ki-kubwa ki-moja ki-li-anguka SG-basket(G7/8) G7-large G7-one G7-PST-fall ‘One large basket fell.’ G=gender class There are, of course, important differences. Firstly, the gender prefix kiin Swahili is a reflex of part of the lexical entry for the noun, and whenever the noun appears, its gender is part and parcel of the syntactic element. In LSE, in contrast, a noun may appear without being localized in space, and only in certain situations, which appear to be discourse dependent, will a location be assigned. Furthermore, in LSE a noun may be assigned one location in a certain stretch of discourse and a different location in another. Secondly, in LSE, not all controllers can admit the form of the marker: body-anchored signs cannot be localized and the association between the sign and the locus 238 Typological perspective of LSE agreement must be achieved by some other means, often by pointing (as described in sections 3.1.1 and 5.1.1). This makes it apparent that the target does not copy directly from the form of the controller, as appears to happen in radical alliterative agreement (see section 2.2.3.3), but rather from some feature that has been associated with the controller. Finally, the form of the marker is affixal in Swahili and a stem alternation in LSE. 6.2.4. Multiple exponence Before closing this section on targets and the way in which they express agreement, I return to the issue of multiple exponence. We have already mentioned multiple exponence earlier in this section as evidence for considering spatial marking a manifestation of agreement rather than some sort of incorporated pronoun. In that case, I referred to the appearance of spatial marking on different elements, namely, the lexical verb and the auxiliary. Another type of multiple exponence occurs when a single target has more than one marker for a given argument. Although the data for LSE are scant in this respect, the possibility that spatial locations are marked not only by the manual component of a sign but also by non-manual features, such as eye gaze or head tilts (as described in sections 3.4 and 5.5), opens up the possibility for such multiple exponence. Given that the multiple articulators allow for simultaneous multiple exponence, this may provide a modalityspecific characteristic of agreement in sign languages, and deserves greater investigation. 1 6.2.5. Summary This section has assessed the targets of agreement in LSE and how agreement is manifest on those targets. The LSE data show that the spatial marking we are considering as agreement, both on agreeing verbs and other elements (under the guise of single argument agreement), occurs on a range of elements that have also been attested for spoken languages. In the verbal domain, the distribution of information between the lexical and auxiliary verb can be characterized as the split/double type, although, in contrast to what is seen in most spoken languages, the auxiliary seems to be specialized for carrying agreement information since it does not inflect for aspect. The means of exponence of this agreement mechanism in LSE is not a form of cliticized 1 An alternative characterization of (manual and non-manual) multiple marking is as a circumfix (Thompson, Emmorey & Kluender 2006). While this reflects the fact that the same information is manifest as two identifiably different parts, the notion of a circumfix (a prefix and suffix) is difficult to reconcile with the manual marking as a stem alternation (and not an affix), and fails to capture the simultaneous nature of the marking. Domains 239 pronoun, contra what has been suggested for other sign languages, but rather a stem alternation through the specification of a phonological feature associated with the controller. This agreement mechanism relies upon a specific use of (locations in) the signing space and does not appear to have an exact parallel in any spoken language, although I have pointed out similarities (and differences) with alliterative agreement. The use of spatial locations is a strongly indexical referencing mechanism that pervades the language well beyond a restricted class of verbs. 6.3. Domains A useful division to make in different domains of verbal agreement is between clause-internal agreement, in which the controller and target are within the same clause, and agreement beyond the clause, as set out in section 2.2.4. We will assess the LSE data within the context of each in turn. Within the clause, LSE creates agreement relationships between the verb and its arguments as has been attested for spoken languages, but also shows agreement marking with various atypical elements, notably with locative arguments, a phenomenon that has been reported for some spoken languages but is nonetheless a rarity. Beyond the clause, LSE displays spatial marking for “pragmatic agreement”, as previously described for other sign languages, and for pronouns, although each appears to involve a slightly different use of the signing space. 6.3.1. Clause-internal agreement In the context of clause-internal agreement, verbs generally agree with their arguments, with a greater tendency to agree with more prototypical arguments, such as subject and object, rather than indirect object or oblique arguments. As we saw in section 2.2.4, the Accessibility Hierarchy (reproduced here in (6)) establishes the types of agreement domain that are prerequisites for others to be present in a language. (6) subject > direct object > indirect object > oblique > genitive > object of comparison The Accessibility Hierarchy (Keenan & Comrie 1977: 66) LSE obeys this hierarchy since it expresses agreement spatially with subject, direct object and indirect object. (The use of localization with spatial verbs could be considered a case of oblique agreement.) Interestingly, agreement for both genitives (possessives) and objects of comparison occurs with the generic agreement auxiliary AUX, as described in section 5.3.1. 240 Typological perspective of LSE agreement Although the data suggest that LSE follows the same patterns that have emerged from cross-linguistic comparisons as far as agreement domains available to the language are concerned, there are some anomalies. Firstly, ditransitive verbs in LSE tend to agree with indirect objects rather than direct objects. However, this tendency has also been reported for many spoken languages, which are said to distinguish between primary and secondary objects rather than direct and indirect objects (see section 2.2.4 for details), and so fits in with established typological patterns. Conversely, a trait common among sign languages, but unusual in spoken languages, is the fact that the object argument seems to be much more salient than the subject argument. This is reflected in the frequency of agreement marker omission for the subject, while the object argument marking is maintained. This issue will be examined further in the context of optionality in section 6.5. Remaining within the clause, agreement in spoken languages has also been described between the verb and non-typical elements, exemplified in section 2.2.4 by possessor raising and copying-to-object formations, in which an argument not central to the verb’s argument structure is marked as if it were such an argument. The data do not reveal that LSE has anything like possessor raising, but a parallel may be found in the ambiguity involved in using space for reference. The examples in (7) show how the same verb may agree with an object/locative, as in (a), or with a person, as shown in (b). (Note that the verb STEAL is a backwards verb, so the object is marked by the first subindex.) LSE (7) a. BIKEX DISAPPEARx xSTEALy ‘The bike’s been stolen (from there).’ b. BIKEX 1STEALy ‘The bike’s been stolen (from me).’ Furthermore, in the case of spatial verbs in LSE, the verb agrees with a locative element. In section 2.2.2, we saw an example from Chicheŵa, a Bantu language that admits locative arguments as the controller of verb agreement, and data of this sort led Thompson, Emmorey & Kluender (2006) to propose that ASL has locative agreement, and that locatives are arguments (rather than adjuncts) of spatial verbs in sign languages. Continuing with agreement with non-typical elements, spatial marking may also be used to connect semantically or pragmatically related concepts for which it is difficult to characterize a specific syntactic relationship. Engberg-Pedersen (1994) identified this type of “pragmatic agreement” for Domains 241 DTS, as mentioned in section 3.2.3, and this also occurs in the LSE data. For example, in a stretch of discourse describing her schooling, a signer uses the space on her left to refer to her school, and the central signing space to refer to high school. Thus, articulating a sign such as FOURTH-YEAR in the left-hand space means ‘(in) fourth grade at school’. In this sense, space is exploited to associate related concepts or ideas, and generally involves the use of broad regions, rather than locations, in the signing space. An explanation of pragmatic agreement will be offered in section 7.2.3 within the formal analysis of spatial agreement proposed in the next chapter. This sort of use of space may stretch over a length of discourse, and thus brings us to domains beyond the confines of a single clause. 6.3.2. Agreement beyond the clause A clear candidate for agreement beyond the clause, already mentioned for spoken languages in section 2.2.3, is pronouns, which display the same features (of the antecedent) that typically show up on agreement markers. That (clause-internal) agreement and antecedent-anaphora employ common underlying mechanisms certainly appears to be the case in LSE, which makes use of the same spatial marking for both. In contrast to the pragmatic agreement mentioned in the previous section, which uses broader, less defined areas of the signing space, pronominal reference relies on more specific points in the space in the same way that localization of verbs does. Thus, any hopes of characterizing clause-internal agreement marking as a clearly demarcated use of space and beyond the clause relations as a more vague use of space (as evidenced by pragmatic agreement) are dashed by the existence of the locus-based spatial marking on pronouns. 2 Furthermore, even when the domain of agreement marking is restricted to a single clause, the process has a strong discourse flavour since markers are assigned for a given stretch of discourse, and not identified in the lexicon. This contrasts with the situation for agreement in spoken languages, such as the gender marking in Bantu languages seen above in example (5), for which the markers depend upon lexically defined features. In the discussion of agreement domains in spoken languages in section 2.2.4, it was mentioned that many of the instances of agreement with non-arguments have been used to show that agreement is conditioned by discourse-level considerations related to highlighting salient animate referents, or even that agreement is essentially a discourse phenomenon. In the case of LSE, spatial marking 2 A determining factor in this respect, which is starting to be investigated in various (Western) sign languages, is the role of specificity in reference. Neidle et al. (2000) compare definite and indefinite reference; Barberà (2014) looks specifically at specificity. 242 Typological perspective of LSE agreement appears to operate at both levels, and serves to mark clause-internal arguments as well as discourse-level relations. It is important to bear in mind that the same formal mechanism may have different functions. Agreement within a specific syntactic configuration may employ marking that is also used for discourse level referencing, and the latter may not be subject to the same restrictions. A final type of agreement relation that has caused much interest in the spoken language literature is long distance agreement (described in section 2.2.4), in which a verb agrees with an argument not in the same clause. With the current LSE data, I have found no examples that could contribute to the debate. Furthermore, since spatial marking can be both intraclausal and interclausal, finding a clear example of long distance agreement in LSE would require showing that a spatially modified verb agrees with a referent from another clause (which is not present in the verb’s clause either as a null topic or a null argument, both of which are possibilities), and that the spatial modification is not some discourse-level mechanism. I leave this arduous task to future research. In summary, the spatial marking used in LSE operates both within and beyond the clausal level, and is conditioned by discourse in the sense that the markers used are created for a given stretch of discourse. Although some uses of space at the discourse level (such as pragmatic agreement) are characterized by a less fine-grained division of the signing space, others, notably pronouns, use space in just the same way as verbs mark their arguments. Nevertheless, form and function should not be conflated, and it may be the case that the same spatial marking is used for various functions, at times syntactically determined and at times governed by broader discourse considerations. The next section looks at the features that are involved in LSE agreement. 6.4. Features and values The type of information, or features, displayed in agreement processes generally falls into three categories: gender, number and person. Grammatical gender is not encoded in all languages, but number and person are taken to be linguistic universals (although some exceptional cases do exist, as explained in section 2.2.5). Additionally, there are other features that have been considered to take part in the agreement process, namely respect and case. In this section, we look at whether each of these features is present in LSE, and if so, how it is encoded. Features and values 243 6.4.1. Gender At first sight, LSE does not appear to code the notion of gender in the verbal agreement process, at least certainly not sex-based gender common to IndoEuropean languages. The different hand configurations used by classifier constructions, which categorize the referent according to its physical properties, could be considered a type of gender system and proposals have been made for classifiers as an agreement mechanism in DGS (Glück & Pfau 1998) and specifically as gender agreement in NGT (Zwitserlood 2003). Such a mechanism would be more similar to the Bantu-style gender agreement mentioned in section 6.2.3 above, in the sense that the gender marker (i.e. the classifier hand configuration) depends upon the semantics of the referent. In terms of the criteria for the gender distinction, these are purely semantic, since the phonological form of a given noun does not affect the class it is assigned to. However, in contrast to the semantic criteria for gender systems in spoken languages, which tend to be based on either sex-based categorization or on animacy, sign languages pay attention to the physical (and mainly visual) properties of the referents of nouns to classify them. Although such an analysis could be applied to LSE (which also makes use of classifier constructions, as described in sections 1.1 and 1.2), our focus here is on the use of spatial marking. This gender system, in contrast, is limited to modifications to the hand configuration, so I shall not pursue an analysis of gender agreement in order to concentrate on spatial marking. 6.4.2. Number Another feature of nominal elements that may enter into the agreement process is number. The most common distinction is between singular and plural, although some languages further distinguish between different degrees of plurality (dual, paucal, etc.). As we saw in section 5.2.1, agreeing verbs in LSE may inflect for number to differentiate between dual, exhaustive and multiple marking, shown in figure 6.1. a b c Figure 6.1 Inflectional forms for marking of plural objects in LSE, as seen from above: a) dual marking; b) exhaustive marking; c) multiple marking. 244 Typological perspective of LSE agreement Additionally, single argument agreement marks number by means of bimanual articulation or reduplication, somewhat similarly to the dual and exhaustive forms of directional agreeing verbs. An important distinction must be made here between verbal and nominal number. In section 2.2.5, we saw that number marking on the verb may reflect both the event semantics of the verb (event number) and the participants (participant number), and as such it overlaps and interacts with both aspect and nominal number as reflected in agreement. In contrast to the expression of (nominal) number in agreement, which displays the value of the number feature of the verb’s argument(s) on the verb, verbal number is inherent to the verb itself and does not involve agreement with anything else. Thus, it is important to try to tease apart verbal and nominal number as expressed on the verb, as only the latter is an instantiation of agreement. Of the markers identified for LSE, which ones reflect verbal number, and which nominal number via agreement? Intuitively, the dual and the exhaustive forms seem to affect the event structure of the verb as they imply at least one iteration of the event, as already alluded to in the discussion of number marking in section 3.2.1.1 based on Wilbur’s Event Visibility Hypothesis. For a lack of rigorous semantic tests to confirm or refute these intuitions, let us turn to the diagnostics suggested by Durie (1986: 357-62), described in section 2.2.5.2, for distinguishing verbal number from agreement markers: i) verbal number operates on an ergative basis, reflecting the number of the most directly affected participant, which is the subject of intransitive sentences (S) or the object of transitive sentences (P), and this may contrast with other marking on the verb (e.g. subject marking, which agrees with S and A). Single argument agreement shows an ergative patterning, since the verb agrees with the internal argument (S or P) and the fact that number marking does not occur for A is not revealing. Directional agreeing verbs, on the other hand, are not intransitive, so it is difficult to make use of the S-P distinction for them. However, considering the A argument (i.e. subject of transitive verbs), there seems to be a clear distinction between, on the one hand, the dual and multiple forms, which may mark A plurality, and exhaustive marking, which cannot. This suggests that the exhaustive may be verbal number since it appears to be limited to S and P arguments. ii) verbal number may mark different values to those marked by agreement, especially when verbal agreement is restricted by some condition. Features and values 251 distinction in LSE. Without this distinction, the person feature does not play a role in agreement or in the referential system of LSE. We now turn to the proposal that location is the feature encoded by agreement in sign languages. Zwitserlood & van Gijn claim that “the only two relevant types of ϕ-features in sign languages are gender and location” (2006: 195-6). (Their claim for gender is based on classifier handshapes as described above in section 6.4.1 and I will not address that issue further here.) Based on the referential use of space, the authors claim that sign languages show location agreement. The term “location” may refer to a morphophonological value (a locus in signing space), semantic feature (locative) or to a specific referent (‘location x’), but it is not clear that any of these things could be considered a ϕ-feature. A ϕ-feature, as pointed out in section 2.2.5, is abstract and categorical in the sense that the different values of the feature provide a means of classifying different linguistic items (typically nominal elements). Therefore, location must be understood as an abstract value that can be differentiated from the actual substantiation of a locus in the signing space; this seems to be Wilbur’s (2013) interpretation when she distinguishes between a geometric point and an actual point in space. If there is some abstract location ϕ-feature, what values does it take? The fact that location might not have a listable set of values is dismissed by Wilbur (2013: 223) as irrelevant to its linguistic status; it is nonetheless the case that other ϕ-features have a small, closed set of values. 5 Another issue for location as a ϕ-feature with respect to the other ϕ-features that participate in agreement is the feature’s properties. Since typical agreement ϕ-features are direct, in the sense that they are associated with prototypical semantics (gender ↔ semantic class; number ↔ numerosity; person ↔ discourse role), this should also be the case for abstract location. Location may be instantiated as a point in the signing space, and this may (or may not) have (locative) meaning, but is this also the case for location as a ϕ-feature? Considering location to be a ϕ-feature appears to be a matter of thinking backwards: when faced with a use of locations in the form of a language, is it licit to postulate that location is part of the abstract grammar of the language? This seems akin to suggesting that English has a sibilant ϕ-feature because plural nouns are marked with a sibilant. Of course, the critical difference is that an argument’s being semantically [±sibilant] makes little sense, whereas having a given [location] value seems to make as much 5 I should point out that Wilbur (2013) does not argue for location to be treated as a ϕ-feature, but rather defends the linguistic status of spatial reference in sign languages. I cite her work here as many of her arguments are germane to the issue in hand. 252 Typological perspective of LSE agreement sense as being second person or masculine. This is a clear effect of modality, since the visual channel makes it possible to employ forms that have a relatively transparent meaning that can be productively exploited. The important question is whether the modality effect goes beyond merely influencing how we analyse the language and actually creates a direct connection between function (ϕ-features) and form (agreement markers) that is unseen and unheard of in spoken languages. In short, the proposal that sign languages make use of a location ϕ-feature is problematic. Firstly, location does not show the typical properties of ϕ-features of having a small set of categorical values. Secondly, it involves conflating the surface form (locations in the signing space) with the underlying abstract feature, which may be something else. While it could be the case that sign languages encode a grammatical meaning that is available to the languages’ medium, I propose another possibility. The ϕ-feature responsible for spatial reference marking, which Zwitserlood & van Gijn (2006) call location and which is indeed realized by means of a location in the agreement process, is an abstract feature that encodes identity but is not intrinsically related to location. This identity ϕ-feature may take different values, which serve to distinguish one discourse entity from another. In this sense, the feature is direct since it is associated with the basic semantics of identity. Concerning the listability issue, in theory the number of values of the feature is unlimited, but in practice it is limited to the feasible number of discourse entities. Admittedly, this does not reduce the number of values to a closed set, but is more limited than any possible location. A given value of this identity ϕ-feature is realized as a location in the phonological form. The breakdowns in indexicality mentioned above, such as stacking or a referent appearing in more than one locus, are due to the correspondence between the surface form and the underlying feature value. This proposal is based on the notion of R-locus (introduced in section 3.1.3), which also makes use of the correspondence with a (discourse-based) referential index in order to resolve the “imperfect” indexicality displayed by the spatial marking. The notion of an identity ϕ-feature attempts to integrate the referential mechanism into the apparatus of agreement, and this proposal will be further developed in chapter 7. To summarize, in spite of the universal presence of the category of person in spoken languages, the use of space for reference in LSE does not employ this feature and as such it is not present in the verbal agreement system. I have assessed the proposal for an alternative ϕ-feature for sign languages, namely location, and have proposed that the abstract feature is not Features and values 253 related specifically to location (even though it surfaces in form as a location) but rather to identity. 6.4.4. Other features: respect and case In addition to the major three features of gender, number and person, agreement systems may also mark other features. As described in section 2.2.5.4, respect and case are marked in the agreement systems of some spoken languages. This section assesses whether the data in this study provide evidence to postulate that these features also exist in LSE, and concludes that there is no evidence for either respect or case. In spoken languages, respect is often marked by means of another feature, typically person (in Italian a third person form is used as a second person respect form, in both singular and plural) or number (in French, second person plural is used as the respect form for second person singular). Other languages have unique forms to mark respect that are not subsumed under another feature, as has been suggested for object honorification in Japanese (see section 2.2.5.4 for details and examples). This distinction occurs in both the pronominal forms and the verbal agreement marking. For LSE, in section 5.1 it was noted that the hand configuration of pronominal forms could change from the more common index-finger point (B) to the flat ] handshape. However, it seems unlikely that this alternation is marking respect as a grammatically encoded category for two reasons. Firstly, the “respectful” form is limited to use with physically present referents, and is not used with non-present referents. This means that the form is limited to deictic contexts and is not possible for anaphoric reference. Such a state of affairs could be expected for second person forms, since the addressee is invariably present, but the unavailability of the respect form for non-present third person reference suggests that the ] handshape is a stylistic polite form for those who are present to perceive it. Secondly, and more crucially, the alternation is possible with first person reference as well as second and third person reference. Since respect marks the perceived social relationship between the speaker and the referent, it makes little sense to have a respect form to refer to oneself. Again, this suggests that the use of the ] handshape is part of a polite register in which certain forms are deemed more appropriate, rather than a grammatically encoded feature. 6 Leaving aside the pronominal system, is there any evidence for a respect feature in verbal agreement marking in LSE? As was pointed out in 6 A parallel can be found in certain varieties of English, in which reflexive pronouns are employed in more formal settings: ‘I myself would like to take this opportunity to thank yourselves for your understanding in this matter’. 254 Typological perspective of LSE agreement sections 3.1.1 and 5.1.1, sign languages (LSE included) may assign referents to different locations in the signing space according to different conventions (Engberg-Pedersen 1993). One such convention involves metaphorical schemes in which the relative social status of referents is reflected by difference in height in the signing space. Thus, for example, a referent perceived to have higher social standing than the signer, such as a doctor, would be assigned to a relatively high location in the signing space, and any verb using spatial marking with the doctor as an argument would incorporate that higher location (Nilsson 2008: 53; Barberà 2012: 115). Could this use of height be considered a respect feature in the agreement marking? Again, as was argued for the pronominal forms, this alternation does not involve a grammatical encoding of a respect feature. Firstly, the marking is not categorical, but involves the use of a continuum along the vertical axis of space. This contrasts with the discreet values displayed by the respect feature (and by ϕ-features in general) in spoken languages. The respect feature in LSE could be characterized as showing finer-grained distinctions in the same way that some languages have multiple number values rather than the simple singular-plural dichotomy. However, this misses the point that the space is being exploited as a continuous scale in much the same way that spatial descriptions in sign language involve an isomorphic mapping that defies a categorical analysis. Furthermore, this use of space is not obligatory and is best described as a convention, evidenced by the fact that it enters into play with other conventions that also influence the choice of location assignment. Thus, for example, in the case of physically present referents, considerations of iconicity have greater weight: a signer talking to a doctor will use a point in space that coincides with the doctor’s location; it would not be possible to mark respect by using a location above the doctor’s real location. As such, this use of height as an indicator of respect in the verbal agreement marking does not form a consistent part of the grammar. It is possible that such a feature could be grammaticalized to form part of the agreement system, but the data indicate that LSE currently does not have a respect feature. Case is often considered a relevant feature in agreement processes, particularly in the nominal domain. As pointed out in section 2.2.5.4, although case is not an inherent feature of a controller, it is closely related to agreement and can be marked together with more typical ϕ-features such as number and gender. In the context of LSE, we saw in section 5.3.3 when looking at verbal auxiliaries that there is an element PERS (derived from the sign PERSON) that behaves very similarly to PRO[bC], a case-marked (dative) pronoun described for ISL (also derived from the sign PERSON). This leads us to ask whether case Features and values 255 is marked in the agreement system of LSE. Even though the PERS element described in section 5.3.3 shows similarities to the ISL pronoun described by Meir (2003; see section 3.3.3), I also pointed out important differences. In contrast to PRO[bC], PERS does not appear to be pronominal, since it may appear with coreferential pronouns. Additionally, the semantic role associated with the marker is not consistent in LSE, since the referent may be AGENT/EXPERIENCER and not the just THEME, as occurs in ISL. Thus, the arguments for considering PRO[bC] to be a case-marked pronoun in ISL do not hold for PERS in LSE. More generally, a case analysis for sign languages has been questioned due to the fact that there is no evidence for case morphology on nominal constituents, the elements that typically carry morphological case (Quer 2011). In conclusion, there is no evidence that case plays a role in the agreement processes of LSE. 6.4.5. Summary This section has assessed which features (and corresponding values) are present in the spatial agreement process in LSE. Of the typical ϕ-features associated with agreement, namely gender, number and person, LSE makes use of number alone. (However, beyond the use of space, gender plays a role in the classifier system in the shape-based classes distinguished by different hand configurations.) Number in LSE can be both nominal and verbal: both may be marked on the verb and I have made a point of distinguishing between the two. From the point of view of agreement, nominal number (of an argument) marked on the verb results from agreement whereas verbal number expresses a feature that is inherent to the verb itself. Nominal number is marked on the verb by means of location(s) and for plurality this translates into movement across multiple locations; verbal number is expressed with movement by means of reduplication of the verb. The lack of a person feature in LSE agreement is unarguably a typological anomaly since person distinctions are attested for all spoken languages. This is a modality effect due to the referential mechanisms available to a spatial language. Alternative proposals have suggested that location is the relevant feature for sign languages, but I have pointed out the need to separate the surface realization (a location in signing space) from the underlying function. I have tentatively proposed that a more basic feature of identity, which serves to distinguish one argument from another, could be at work in LSE agreement, and this proposal will be taken further in section 7.1. Even with this radical difference in the set of features available to agreement in LSE, the system still fulfils the generalization observed by Moravcsik (1978: 369) mentioned in section 2.2.5.4: whatever features are available to the 256 Typological perspective of LSE agreement agreement system of a language will also be available in the pronominal system. The identity feature I propose here is also used by the pronominal system since the indexical points also avail of the locations that are the phonological manifestation of the identity feature. 7 As such, LSE follows the established patterns for the behaviour of the features involved in agreement. 6.5. Conditions Conditions are factors that are not realized in the agreement process, but which may determine how (or whether) agreement takes place. In section 2.2.6, we saw how animacy and topicality affect agreement in various languages, such that agreement will only occur if an argument is [+animate] or a topic, for example. In the light of the properties of agreement described in the sign language literature, especially the uneven distribution of agreement across the verbs of a given language, we considered in section 3.2.1.3 possible conditions and prerequisites that could explain the attested limitations on agreement. This discussion was framed within the characterization of agreement on directional agreeing verbs, which display a specific set of properties: agreement occurs on (di)transitive verbs of transfer, and arguments must be [+human], [+animate] or a possessor of some sort. However, I have argued that agreement in LSE should not be restricted to this type of agreeing verb and that there is a more general process involving an association between a referent and a location in signing space. This process is not limited to twoor three-place predicates and occurs with individual arguments as single argument agreement (sections 3.2.3 and 5.2.3). In this section, I adopt this broader perspective on agreement to examine the conditions that operate on this spatial agreement mechanism in LSE. This discussion will lead to the issue of optionality in LSE agreement. Animacy is a well attested condition on agreement in many spoken languages, with less animate arguments being less likely to trigger agreement (see section 3.2.1.3 for examples). Various proposals have suggested that agreement is restricted to animate arguments in different sign languages (for ASL, Mathur 2000: 212; for Korean Sign Language, Hong 2008: 170). However, even within the domain of (two-place) agreeing verbs, this condition does not hold cross-linguistically: in section 3.2.1.3 we saw example (14) from LSQ of an agreeing verb with inanimate arguments, and in section 5.4.1, LSE example 7 Furthermore, the proposal for classifier handshapes as some sort of class or gender agreement (section 6.4.1) also fits into this generalization. If classifier proforms are considered as another type of pronominal reference, then the gender feature is also present in the pronominal system. Conditions 257 (23). Furthermore, expanding the notion of agreement marking to single argument agreement makes this condition less tenable: examples from ISL (see (22) in section 3.2.3) and LSE (see (10) in section 5.2.3) demonstrate that inanimate arguments such as ‘stick’ or ‘exam’ may trigger agreement. Furthermore, the arguments of the so-called spatial verbs are inanimate. Padden’s (1983/1988) tripartite classification of verbs does not characterize the spatial modification of spatial verbs as an agreement process; I now reassess this distinction. In section 3.2.1.4, I questioned the distinction between agreeing and spatial verbs: various authors have already proposed models that do not maintain such a distinction (for ASL, Janis 1992; for Libras, Quadros 1999). I will draw upon two separate observations to support the claim that both agreeing and spatial verbs manifest the same agreement process. Quadros & Quer (2008) suggest that agreement occurs in both categories of verb, which may agree with either locative or personal arguments. 8 This coincides with the idea mentioned above in the section on the controllers of agreement (6.1) that sign languages have locative agreement (Thompson, Emmorey & Kluender 2006), and locative arguments form part of certain verbs’ argument structure. Therefore, sign language verbs agree with their arguments, some of which are locative, and this is achieved by means of the same spatial mechanism of assigning the referent to a location in signing space. This use of space for agreement should be distinguished from the exploitation of space for isomorphic mappings typically employed in spatial descriptions. In the case of locative arguments, the former tends to occur within the context of the latter, such that the locations assigned to arguments occur within a spatial map. By their semantic nature, locative arguments are places, so they occur in discourse contexts in which location is likely to be relevant, and spatial information is therefore represented (by means of isomorphic mapping of the signing space). However, the two uses of space are independent, as is evident from the following two cases that highlight the distinction. Firstly, non-locative arguments may occur in isomorphic spatial mappings. This occurs in the case of present referents, in which the locations assigned to referents correspond to each one’s real-world location. Nonpresent non-locative referents may also occur in a spatial map, motivating the use of a higher location in signing space in examples of the type ‘I asked the tall man’ (cf. Liddell 2000). Secondly, locative arguments may appear in 8 Quadros & Quer maintain the presence of a person feature, which I have rejected above in section 6.4.3. My analysis makes use of an alternative identity feature, but the relevant point here is that arguments may be locative. 258 Typological perspective of LSE agreement contexts in which relative location is not relevant, and thus an isomorphic mapping is not used. This dissociation is apparent in LSE due to the fact that isomorphic mappings of large physical distances tend to occur in the vertical plane: a signer describing where she has studied may place Bilbao, Barcelona and London at different points on the vertical plane to show their relative positions, as if she had a map in front of her. However, if the geographical location of each place is not relevant, each referent may be associated with a point on the horizontal plane. In this context, any verb that agrees with those locations (‘I moved from Barcelona to London’) is not susceptible to the (iconic and scalar) properties of topographic space. Indeed, the properties of spatial verbs that Padden (1983/1988) described (and which seemed to set them apart from agreeing verbs) are in fact properties of isomorphic, topographic space in which spatial verbs tend to (but do not necessarily) occur. In summary, since the spatial agreement marking in LSE operates in a similar way for agreeing verbs, single argument agreement and spatial verbs, the arguments are not restricted to a specific semantic property, but take in animates, inanimates and locatives. As such, there does not appear to be a condition related to the semantic category of the arguments of agreement. However, although arguments of different types may trigger agreement, the issue remains that agreement in sign languages is highly optional. As described in section 3.2.1.2, agreement marking may be omitted for one of the arguments (agreement marker omission), or may be completely absent. In the first case, the subject argument is not marked, while marking for the object argument is preserved. As mentioned above in section 6.3.1, this appears to be related to a general tendency for the object argument to be more salient in sign languages, and requires further investigation to determine whether this could be a modality driven effect. Nevertheless, although the subject argument is not marked by the verbal inflection, alternative mechanisms could substantiate the agreement process such as a role shift operator, which may be characterized as another type of agreement (Herrmann & Steinbach 2012). Alternatively, the lack of subject marking may be due to specific types of reference: for example, null arguments (with null agreement marking) may be used to mark impersonal reference (Kimmelman 2015). The complete absence of agreement marking on a verb that might potentially agree, on the other hand, suggests that the agreement relation is completely absent. Since agreement marking is achieved by the use of the locations in signing space associated with the corresponding referents, this issue is closely related to whether or not location assignment occurs. If locations are not assigned, they are not available to the agreement Canonicity 259 mechanism. Whether or not a location assignment occurs seems to depend upon discourse considerations to do with the presence of other referents and whether the predicates give rise to the need to track those referents. For example, a semantically reversible sentence (‘Bea likes you’) may require (spatial) means to clarify the semantic/syntactic role of each argument, whereas this is not so for a non-reversible proposition (‘Bea likes maths’). Equally, even when a verb only agrees with a single argument (as in the case of single argument agreement), the presence of competing referents in the discourse may force the use of space to distinguish between them. In this sense, the use of spatial reference (and hence of the agreement mechanism that depends upon it) appears “as required”, perhaps along lines similar to the use of explicit pronouns in pro-drop languages. This “only if needed” nature of LSE agreement is an unusual property, since agreement is generally taken to be a basic, automatic mechanism in a language. The optional nature of spatial agreement will be returned to in section 7.1.3 since it represents an important challenge to a syntactic account of agreement in LSE. To conclude this section on conditions, agreement in LSE, understood as a basic mechanism that uses locations in the signing space as markers for arguments, is not subject to semantic conditions. In contrast, the widespread optionality of LSE agreement highlights the fact that discursive and pragmatic considerations determine the appearance of spatial marking, and as such represent a condition on agreement that merit further investigation. 6.6. Canonicity Having described the workings of spatial agreement in LSE in some detail, and having assessed how the different aspects (controllers, targets, features, etc.) measure up against what has been described for spoken languages, I now turn to an evaluation of how agreement-like this mechanism is. In order to do this, I use Corbett’s (2003b, 2006) notion of canonicity. As described in section 2.2.7, the notion of canonicity is essentially a means of defining prototypical agreement based on a set of general principles that characterize the phenomenon and on consensus (in the literature and across different theoretical frameworks) about what counts as agreement. Thus, the properties of prototypical agreement conform to general characteristics of agreement (e.g. “agreement is redundant rather than informative”) and are those that are accepted by most linguists as qualifying as a case of agreement (e.g. “the domain is local”). Prototypical agreement is defined in terms of a series of specific criteria (e.g. “marking is bound”) and occupies the centre of the space of possible agreement-like phenomena. The remaining space is mapped by 260 Typological perspective of LSE agreement the less prototypical alternatives for each criterion (e.g. “the domain is nonlocal”, “marking is free”). C-1: controller is present > controller is absent C-2: controller has overt expression of agreement features > controller has covert expression of agreement features C-3: consistent controller (all targets take the same value for a given feature) > hybrid controller (targets take different values for a given feature) C-4: controller’s part of speech is irrelevant > controller’s part of speech is relevant C-5: marking is bound > marking is free C-6: marking is obligatory > marking is optional C-7: marking is regular (affixal) > marking is suppletive C-8: marking is alliterative (marker on all targets is the same and identical to formant on controller) > marking is opaque (marker changes from target to target and is not identical to formant on controller) C-9: marking is productive (applies to all members of a category) > marking is sporadic (only appears on some members of a category) C-10: target always agrees > target agrees only when controller is absent location assignment occurs C-11: target agrees with single controller > target agrees with more than one controller C-12: target has no choice of controller > target has choice of controller C-13: target’s part of speech is irrelevant > target’s part of speech is relevant C-14: domain is asymmetric > domain is symmetric C-15: domain is local > domain is non-local C-16: domain is one of a set > single domain C-17: feature is lexical > feature is non-lexical C-18: features have matching values > feature values do not match C-19: no choice of feature value > choice of feature value C-20 no conditions > conditions Table 6.2 Criteria for canonical agreement. The symbol > means “is more canonical than”. Adapted from Corbett (2006: 10-27). For each criterion, the behaviour of spatial agreement marking in LSE is indicated in boldface and shaded background. The original formulation of C-10 is adapted to the case of LSE and the change indicated by strikethrough and italics (see text for details). Canonicity 267 The final criterion for canonicity (table 6.2: C-20) refers to conditions: canonical agreement has no conditions. Once a target, controller, domain and features have been defined, this should be enough to fully specify the agreement relation. However, for LSE, we have seen that the presence and expression of agreement is subject to various conditions: notably, discursive and pragmatic considerations determine whether space will be used for reference tracking, and also the physical presence or visibility of the referents in the communicative context. As such, agreement in LSE is not canonical in this respect. In summary, this assessment of spatial agreement in LSE shows that on the basis of most of the criteria the process is more canonical than not: 15 out of 20. It is important to bear in mind that most spoken languages also present varying numbers of non-canonical properties. As a comparative, a precursory evaluation of verbal agreement in spoken Spanish suggests that the process is canonical according to 16 of the criteria, but has less canonical behaviour for four criteria (C-1, C-2, C-8 and C-18). Nevertheless, these criteria serve as guidelines, and do not furnish us with an index of canonicity: as Meatloaf so aptly points out, “two out of three ain’t bad” as far as scorekeeping goes, but more important is the nature of the measures being applied. To avoid trying to make sense of a statement of the type “LSE agreement is 75% canonical”, I turn to the general principles that underlie the criteria for canonicity to provide a broader perspective (Corbett 2006). 6.6.2. Applying Corbett’s general principles to spatial agreement in LSE Three general principles underlie the criteria for canonicity. These principles concern redundancy, syntactic simplicity and the morphological expression of agreement (Corbett 2003b, 2006). Each criterion is motivated by one (or more) of these principles, and these dependencies are shown in table 6.3. The table also indicates whether spatial agreement in LSE is canonical for a given criterion by means of a tick or cross. The first general principle states that canonical agreement is redundant rather than informative. This principle groups together criteria 1, 2, 10, 17, 18 and 19 (Corbett 2006: 27). As can be seen in table 6.2, spatial agreement in LSE meets just half of these criteria, which indicates that the system displays a relatively low degree of redundancy. This lack of redundancy stems from two aspects of the referential system that the agreement mechanism exploits: firstly, the marking is strongly indexical (rather than lexical) in nature (cf. C17) and so in some sense the marker is adding rather than just repeating something (which would be more redundant). 268 Typological perspective of LSE agreement I Canonical agreement is redundant rather than informative. II Canonical agreement is syntactically simple. III The closer the expression of agreement is to canonical inflectional morphology, the more canonical it is as agreement. C-1  C-2  C-3  C-4  C-5  C-6   C-7  C-8  C-9  C-10   C-11  C-12  C-13  C-14  C-15  C-16  C-17  C-18   C-19   C-20  Table 6.3. The general principles for canonical agreement, and the criteria associated to each principle. The ticks and crosses indicate whether or not spatial agreement in LSE is canonical according to a given criterion. The lighter shading of C-6 for general principle II indicates that this criterion is only partially associated (see text for details). The second property of the referential system that reduces redundancy is the “only if needed” quality mentioned in section 6.5 (and reflected in C-10). If spatial reference were obligatory whenever a referent is introduced, this would create a much more redundant system. However, LSE does not always anchor reference to space (and thence use spatial agreement) but only whenever the need arises. The second general principle characterizes canonical agreement as syntactically simple, and takes in criteria 3, 4, 10, 11, 12, 13, 14, 15, 16, 18, 19 and 20 (and partially 6) (Corbett 2006: 27). According to this principle, Canonicity 269 agreement that can be described by straightforward rules is more canonical. In this respect, spatial agreement in LSE is broadly canonical (especially from the point of view of the consistency of controllers, targets, their marking and features), but fails on two related counts. The optionality of the mechanism means that agreement does not consistently occur (cf. C-6 and C-10) and this complicating factor is (at least partially) expressed in terms of conditions on the agreement process (cf. C-20). The third general principle relates to how agreement is marked: the closer the expression of agreement is to canonical (i.e. affixal) inflectional morphology, the more canonical the agreement mechanism. This motivates criteria 5, 6, 7, 8 and 9 (Corbett 2009: 27). As far as the marking of spatial agreement in LSE is concerned, it behaves like canonical inflectional morphology with the peculiarity that the marking is a stem alternation. Still, this idiosyncrasy may even be considered to be canonical in the context of the simultaneous templatic morphology of signed languages (Sandler & LilloMartin 2006). However, once more, the optional nature of the process makes spatial agreement in LSE less canonical: the fact that marking may be omitted makes it less like (canonical) inflectional morphology. In light of these general principles, the anomalous or non-canonical properties of spatial agreement in LSE can be seen more clearly. What becomes apparent is that the factors that pull the agreement system away from canonicity derive from two general areas: on the one hand, the related issues of redundancy and optionality, and, on the other, the nature of the referential system exploited by agreement in LSE. The optional character of agreement in LSE makes the system less redundant since it often will not appear if it is not required. This raises the question of what determines the optionality of the system: when can or must agreement appear, or not? I have already indicated that pragmatic and discourse considerations play a role in this respect, but further study is required to establish a more precise characterization of the factors that have an effect on the appearance of agreement in LSE. To this end, a corpus-based study could provide valuable evidence to identify what causes agreement to occur or not. A study of an Auslan corpus established that verbs modify to indicate their arguments relatively infrequently (less than half of the tokens) (de Beuzeville, Johnston & Schembri 2009). The authors suggest that the availability of other strategies (such as constituent order, information structure and constructed action) may influence whether or not spatial agreement occurs but it remains to be seen if the choice of one strategy or another is systematically conditioned by discourse, pragmatic or syntactic factors. 270 Typological perspective of LSE agreement Secondly, the fact that the marking is spatial, and based on a spatial referential system, comes to bear on the nature of agreement in LSE. The agreement process in LSE involves adding a spatial label to the controller of an agreement relation, rather than exploiting an existing lexical feature (such as gender). This means that both controller and target are marked with the feature, possibly weakening the asymmetric directionality of the relation. Nevertheless, the use of spatial reference also has canonical aspects. Corbett (2006: 24) points out that features based on formal assignment are more canonical than those for which assignment is semantically based. Although in LSE the semantic and formal values of a locus tend to coincide, the important observation is that the locus need not be assigned based on semantics. This was mentioned in the discussion of the distinction between spatial and agreeing verbs (in section 6.5); the point is that spatial loci may be used exclusively for reference with no semantic spatial meaning. Thus, although the use of loci can be (non-canonically) semantic, it is possible for the loci to serve a merely referential function. This exercise in assessing spatial marking in LSE in terms of canonicity does not provide a definite classification of the phenomenon as “agreement” or “not agreement”. Although it has been claimed that the issue of (non-)canonicity has been overstated (Quer 2011: 196) and it is true that the agreement systems of many languages show non-canonical properties (recall that all pro-drop languages fail C-1), I agree with Corbett (2006: 27) that “it is more important to understand agreement and its related phenomena than to draw a precise line” between agreement and other phenomena, and considerations of canonicity provide a useful perspective for observing and analysing the matter. 6.6.3. Other evaluations of the canonicity of sign language agreement Before closing this section on canonicity, I look at the two other attempts to apply this concept to agreement in a sign language, both of which focus on ASL (although the first paper also presents data from other sign languages). Mathur & Rathmann (2010) provide a somewhat cursory evaluation of sign language agreement based on Corbett’s criteria for canonicity. Their conception of agreement in sign language is limited to directional agreeing verbs. 9 They evaluate all of Corbett’s criteria (with the exception of C-20, which they do not mention), and most of their verdicts coincide with mine, so 9 Mathur & Rathmann (2010) distinguish between double agreement (i.e. with subject and object) and single agreement (i.e. with the object). Note that the latter is not what I have called single argument agreement but is a directional agreeing verb with agreement marker omission. Canonicity 271 I will focus on the points of divergence. Concerning targets, there are two differences, both of which concern the issue of optionality: for C-6, Mathur & Rathmann consider that the marking is obligatory, whereas I have claimed that it is optional given the phenomenon of agreement marker omission; and for C-10, they maintain that the target always agrees, whereas I pointed out that this only occurs when spatial loci are assigned. Unfortunately, they do not provide much explanation or clarification for their decisions, so it is not possible to clear up this difference. The other discrepancy comes from the criteria for features: for C-17, they judge features of agreement to be lexical; for C-18, they hold that the feature values do not match. Their underlying analysis of agreement (as involving the features of person and number) is what leads them to differ in these criteria. Additionally, while they consider role shift to be a reason for mismatches between features (e.g. first person verbal forms for non-first person controllers), I considered that such cases constituted a different domain that motivates a different expression of a feature value. In summary, despite minor differences due to different characterizations of the agreement process (and its scope), Mathur & Rathmann’s assessment of the canonicity of sign language agreement coincides to a great extent with that of the LSE data presented here. In their discussion of the linguistic status of sign language agreement, Lillo-Martin & Meier (2011) dedicate a section to the non-canonical properties of directionality as agreement. However, they do not frame the discussion in terms of Corbett’s criteria, but rather in terms of the issues discussed in the sign language literature, namely, verb classification and the prominence of object (over subject) marking. This leads to an interesting analysis of agreement in sign language (and highlights topics that have cropped up in the discussion of the LSE data here, notably, optionality) but does not provide an evaluation from within an independent framework that can provide a systematic means of gauging how and where a given agreement system “misbehaves”. Additionally, Lillo-Martin & Meier’s discussion is limited to directional agreeing verbs and does not contemplate spatial marking for other targets, such as single argument agreement. 6.6.4. Summary Spatial agreement in LSE shows many properties of canonical agreement as defined by Corbett (2003b, 2006), and this broadly coincides with the only other assessment that has systematically evaluated each of the criteria for agreement in sign language based on data from other languages (Mathur & Rathmann 2010). Furthermore, assessing the properties of agreement in terms of wider reaching general principles has provided a way of analysing spatial 272 Typological perspective of LSE agreement agreement in LSE to highlight its peculiarities as an agreement process. Where the phenomenon deviates from canonical behaviour, this is in large part due to two factors: the optionality of the process and the spatial reference mechanism that it makes use of. 6.7. Discussion and conclusions This chapter has analysed the LSE data on spatial agreement from the point of view of the typological approach set out in section 2.2. This examination has shown that agreement in LSE, understood as the use of space to mark arguments on verbs and other lexical categories (such as adjectives, determiners, and so on), falls within the boundaries of the phenomenon of agreement as described for spoken languages. Importantly, the phenomenon I have assessed is not limited to a subset of verbs that show subject and object agreement (i.e. agreeing verbs) but takes a wider view of agreement as a basic process at use in the language. The first part of this assessment involved reviewing the different elements that play a role in agreement: controllers, targets (including means of exponence), domains, features and conditions (sections 6.1-6.5). In each case, spatial agreement in LSE is comparable to the findings for spoken languages (although some unusual characteristics, to be discussed below, also turned up). This conformity was also confirmed by the second phase of the analysis, which applied Corbett’s (2003b, 2006) notion of canonicity to the LSE data to see how agreement-like this mechanism is (section 6.6), both from the point of view of the individual criteria (section 6.6.1) and the general principles that underlie the notion of canonical agreement (6.6.2). On both counts, the spatial agreement mechanism in LSE fares well, and qualifies as more agreement-like than not when viewed through the lens of canonicity. As pointed out during this analysis, it is important to bear in mind that the end result of this analysis is not the most interesting finding (nor, in the final reading, can it provide us with a conclusive categorization of spatial agreement as an instance of agreement). Rather, the process of analysing the data in this way has led to a refinement in how we conceptualize this spatial mechanism in LSE and has also brought to light those properties that are unusual or anomalous. In terms of refinements, this chapter includes several analyses of the LSE data that provide a clearer, more motivated description of the general phenomenon. When looking at the means of exponence of this agreement mechanism (section 6.2.3), I assessed previous claims that the spatial marker is some sort of pronominal affix and applied various criteria related to the properties of Discussion and conclusions 273 agreement markers and pronominal affixes in order to categorize spatial inflection in LSE as one or the other. This evaluation pointed in the direction of agreement marking, and also helped to strengthen the case for treating single argument agreement as a valid manifestation of this agreement mechanism. It also highlighted the difference between a pronoun and a point in space: when the two are confounded, what is essentially a phonological feature may erroneously be considered a clitic. I presented a characterization of this spatial marking in LSE as a stem alternation, a mechanism also attested for spoken languages. The discussion of the number feature brought up the distinction between verbal and nominal number (section 6.4.2). Given that LSE has various mechanisms for marking plurality on the verb, I applied a series of diagnostics to ascertain whether this inflection was a reflex of the numerosity of the verb’s arguments, or of verbal number. The results confirmed the initial intuition that the reduplicative process present in the distributive form marks verbal number, while argument number is marked through the use of space (whether that be the arc movement of the multiple marker, or the two distinct locations in the dual marker). The final refinement came about in the context of the conditions that operate on this agreement process (section 6.5). When analysing the common claim that agreement in sign language is restricted to human or animate arguments, I pointed out that considering spatial marking as the agreement mechanism removes this constraint since single argument agreement shows a much wider variety of arguments (and the data show that even some agreeing verbs are not subject to this semantic restriction). The proposed semantic restriction may be relevant for (some) agreeing verbs, but not for the agreement mechanism as a whole. Furthermore, the inspection of the semantic properties of verbal arguments led to a reassessment of the difference between agreeing and spatial verbs: I proposed that spatial verbs have locative arguments but the agreement mechanism is the same for both types of verb (contra Padden 1983/1988). What can make a fundamental difference in the use of space is isomorphic spatial mapping, which involves strong iconic motivation. The fact that spatial verbs (with their locative arguments) frequently exploit isomorphic mappings confounds this distinction. However, I provided examples that demonstrate that locative arguments do not necessarily imply an isomorphic use of space. Thus, space may be used in sign languages in different ways, but the mechanism of interest here – the marking of arguments via locations – has many of the hallmarks of agreement. 274 Typological perspective of LSE agreement Although this process of spatial marking is a strong candidate for agreement based on this assessment in terms of cross-linguistic data from spoken languages, the analysis also pointed out areas where LSE is doing something out of the ordinary as far as agreement is concerned. The most striking – and typologically extraordinary – aspect of spatial agreement is the fact that it does not make use of a person feature. Although a person distinction has been argued for in other sign languages, the data do not support upholding this feature for LSE. This makes LSE extremely unusual as a language. I argue that the lack of the person feature stems from the referential system employed by the language. In turn, this referential system is based on the use of space and the lack of person marking is a consequence of the possibilities afforded by a spatial reference mechanism. As an alternative to the person feature, I have proposed an identity feature, based on the notion of R-locus, and this proposal will be developed in the next chapter. Another anomaly of spatial agreement in LSE that cannot be avoided is its pervasive optionality. This issue has been mentioned at many points in this thesis (and will continue to rear its head), and is one of the main causes for the non-canonical traits of agreement in LSE. This aspect of sign language agreement also seems to be the cause of much scepticism as to its status as agreement, especially in the spoken language literature, and I will mention two different references in this respect. Firstly, in his comprehensive study of agreement, Corbett (2006: 264), on whose work much of this thesis has so heavily leant, mentions sign languages in a footnote and states that the process described as agreement for these languages does not seem to have the “systematic covariance” to be considered agreement. Even though I could try to claim that Corbett’s dismissal of sign language agreement is not relevant to my analysis since I consider spatial agreement per se and not just agreeing verbs, the underlying issue that his comment alludes to is not diminished: optionality. The second reference is Cysouw’s (2011) reply to Lillo-Martin & Meier’s (2011) landmark paper on the linguistic status of sign language agreement. Cysouw states that Lillo-Martin & Meier convincingly show that pointing and directionality in sign languages is comparable to person marking in spoken languages, but that this does not allow them to take the further step of claiming that this process is agreement. Clearly this depends on the definition of agreement, which Cysouw goes to great lengths to trace through its historical evolution, and the underlying problem appears to be the fact that this person marking mechanism does not represent a “systematic covariance” of linguistic expressions. Again, I could argue that my conception of agreement as a spatial marking mechanism is much more systematic in that Discussion and conclusions 275 it applies to a wider variety of verbal agreement in addition to agreement in the nominal domain. However, I am still left holding the baby as far as optionality is concerned. In the section on conditions (6.5) I went some way to addressing the issue of optionality by suggesting that it is due to discursive and pragmatic conditions on agreement: only when the right circumstances hold will agreement take place. This still leaves unanswered the question as to how the agreement system would handle this endemic optionality: the formal account developed in the next chapter attempts to explain this in terms of default values. Before closing this chapter on the commensurability of spatial agreement in LSE and agreement in spoken languages, we can now address the second of the research questions from chapter 1: Are the spatial mechanisms employed by LSE comparable to the agreement mechanisms in spoken languages? This chapter has provided a detailed analysis of the various spatial mechanisms that I earmarked as being likely candidates for agreement in LSE and described in chapter 5. This appraisal indicates that spatial marking in LSE shows strong parallels with agreement in spoken languages and certainly appears to fall within the limits of the phenomenon as manifest across spoken languages. The assessment of the canonicity of LSE spatial agreement also yielded a favourable result, suggesting that this mechanism is a relatively canonical instance of agreement. As Mathur & Rathmann (2010: 196) point out in their evaluation of canonicity of sign language agreement, “other approaches to verbal agreement in signed languages may interpret Corbett’s criteria for canonical agreement differently” and it is worth bearing in mind that this diagnostic tool depends greatly on the prior conceptualization of the process under examination. Indeed, some discrepancies between Mathur & Rathmann’s canonicity analysis and my own can be traced back to their restriction of agreement to agreeing verbs and my inclusion of related spatial phenomena. In sum, spatial agreement in LSE, understood as the spatial marking of one element to mark covariance with another, shows enough similitude to the typological concept of agreement used for spoken languages, for both to be treated as manifestations of the same linguistic process. Considering these phenomena from different modalities as the same is both meaningful and useful for arriving at a better understanding of how and why a language creates relations between its elements. (This is not to say that there are no differences, and this section has also brought these to the reader’s attention.) The findings of this chapter also make it possible to build upon the preliminary answer to the first research question offered at the end of the 276 Typological perspective of LSE agreement previous chapter. The fine tuning of the characterization of spatial agreement in LSE from the analysis in this chapter puts us in a better position to compare the LSE data with what has been described for other sign languages. In many ways, the characterization of LSE agreement I have developed has diverged from the situation for other sign languages as laid out in chapter 3. Spatial agreement in LSE takes in a broad range of phenomena, well beyond the domain of agreeing verbs, and does not involve a person feature, which is not what we saw for other sign languages. Some of these differences may be due to genuine intra-modal cross-linguistic variation: the differences in the data for pronominal forms between LSE and ASL, for instance, suggest that some sign languages may make use of a person feature while others do not. However, other differences may have more to do with the evolution of the theory of agreement that I have developed for LSE: only a reanalysis of the data for other sign languages will reveal whether this model can be applied to explaining spatial agreement in those languages as well. In the interim, the data from LSE and other sign languages offer enough similarities to suggest that a similar mechanism is at work and enough differences to make comparative studies worthwhile. The next chapter continues to analyse the data for this study, and provides a formal account of spatial agreement in LSE. Location, identity and locating identity 283 dedicated Num projection has been proposed for number and an independently motivated nominalizing n projection for gender/word class. I will deal with each in turn. (2) The Num projection is a midlevel functional category analogous to T in the verbal domain (Ritter 1991; Carstens 2001) and provides the structural position for the number feature. Ritter (1993) further claims that for Romance languages the gender feature is also located at Num, and not on N, the noun head, as is the case for Hebrew. This is based on observation of the inflectional morphemes in these languages, and it is thus an empirical question for a given language (or language family) whether gender appears on Num or not. If the identity feature occupies Num, this means that it is in the same functional head as the number feature (in much the same way that number and gender are in Romance languages according to Ritter’s proposal). This could provide predictions about the position of number and identity relative to other elements that could be tested empirically. However, the fact that much morphology in sign languages is simultaneous rather than sequential makes it difficult to pinpoint the relative position of these features in the structure based on surface form. Furthermore, both the number and the identity feature may be manifest spatially (and thus at the same time), obscuring even more how each feature contributes to the final form. The LSE 284 Formal analysis of LSE agreement data in this study do not provide any clear evidence either for or against the person and identity features occupying the same structural position. 5 Alternatively, the host for the identity feature could be n. This projection immediately dominates the NP and is effectively what makes the NP function as a noun in the syntactic structure. As we saw above, it has been claimed that the gender (or, more precisely, word class) feature is located in this projection (Kihm 2005). Given that the identity feature is assigned to the nominal, it is possible that it occupies the head of a nearby (i.e. immediately dominating) projection in order for the identity feature to affix to the NP, thus forming a “syntactically complete” nominal, that is, an nP. However, two considerations point away from this possibility. Firstly, word class/gender provides a means of categorizing lexical items according to some underlying semantic classification, and this is not what the identity feature appears to do. The second consideration concerns the notion of referential index and is also relevant when evaluating N itself as a potential host for the identity feature. So we will now examine this possibility. As mentioned above, it seems unlikely that the identity feature is part of the lexical entry of a given noun since the same noun may appear with different identity values (i.e. referring to different entities). Nevertheless, Baker (2003, 2008) defines nouns as lexical categories that have a referential index, and it is this referential index that the identity feature appears to pick up on. Indeed, Baker speculates that bearing a referential index depends on more fundamental underlying criteria of identity (2008: 31-33). This link between identity and the (modality-independent) referential index is an insightful connection and provides support for my proposal for an identity feature, but I question the association between the noun itself and the referential index. Baker mentions the principle that an XP must have a referential index in order to have intrinsic ϕ-features (values for person, number and gender). For Baker, this XP would be an NP; for a proponent of an nP account sketched in the previous paragraph, it would be an nP. Nevertheless, as we saw in section 7.1.1, the ϕ-features associated with a nominal are distributed throughout structure that goes beyond the NP (and the nP). As such, it is the entire DP that contains the ϕ-features and so, in keeping with this principle, it must be the DP that has the referential index. Independent support for this idea comes from work on determiners that assumes that the determiner licenses the appearance of a noun as an argument (Longobardi 1994). Evidence that DPs can be arguments whereas 5 Examining how classifier constructions interact with number and the use of location may provide a useful means of probing this issue and I leave the matter for future research. Location, identity and locating identity 285 NPs cannot comes from contrasts such as (3) in Spanish. In (3a), a determinerless NP (hermano de Amaia) may occupy a predicate position, whereas in (3b) a full DP (with the definite article el) is required in an argument position and the sentence is ungrammatical without the article. Furthermore, support from LSE itself comes from the fact that whole sentences can be localized and thus operate as arguments. (This link between localization and nominalization will be developed in section 7.2.3.) Spanish (3) a. Itzal es hermano de Amaia ‘Itzal is Amaia’s brother.’ b. A Itzal le gusta *(el) hermano de Amaia ‘Itzal likes Amaia’s brother.’ Since the full DP carries the referential index (and the ϕ-features), there is no need for the identity ϕ-feature to be restricted to the N (or n) position. Furthermore, the very fact that the full DP is associated with the referential index makes D a much stronger candidate as the host of the identity feature than N (or n) is. The connection between location and the D head had been touched upon in previous work. In her work on LIS, Bertone (2007) proposes that space features are hosted on D. Building on this analysis Brunelli (2011) places location assigning points also in the head of the DP. 6 We saw above that the person feature is situated on D for spoken languages: since here I claim that LSE has an identity feature and no person feature, a parsimonious solution would be for these complementary features to occupy the same position. Indeed, there is a lot in common between the features of person and identity, since both take values that depend upon the context of utterance. The person feature does this by dividing referents according to their role as participants in the discourse (speaker, addressee, etc.) while the identity feature distinguishes referents that appear in the discourse. One possible problem with this proposal is the optionality of the use of location. Abner (2012) points out that a non-obligatory determiner that may or may not appear flies in the face of the idea that languages cannot exhibit free variation between the presence or absence of a determiner for nominal arguments. Thus, if a language has a lexical determiner with a certain meaning, it must use that determiner to express that meaning (Crisma 1997, 6 The idea that location features are hosted on D is exploited by Pfau & Steinbach (2013) in their analysis of the grammaticalization of the sign PERSON in DGS. 286 Formal analysis of LSE agreement cited in Longobardi 2001: 584). However, the claim here is not for a specific lexical determiner in D (along the lines of MacLaughlin’s (1997) analysis for prenominal points in ASL, which is what Abner was arguing against) but rather for the presence of a feature that enters the syntactic structure at a specific place. 7 Furthermore, the optionality of a point is a separate issue from the optionality of the identity feature: this is demonstrated by the possibility of localization, in which an identity feature is present and expressed, but a point may (or may not) be absent. 8 The apparent optionality of localization doubtless complicates the issue but it is not specifically problematic for the identity feature being hosted on D. 7.1.3. Optionality of the use of space In the above discussion of the identity ϕ-feature, the issue of optionality cropped up several times. The fact that location is not always used raises an important question: is the underlying identity feature optional, or, alternatively, is the feature present but (sometimes) phonologically null? Given that the identity feature reflects a fundamental underlying concept, it seems more likely that it is present but may give rise to a phonologically null realization. Other ϕ-features show similar behaviour: in section 2.2.5.2, we saw how number may have a neutral value such that the corresponding form is ambivalent with respect to numerosity, and often this neutral value is expressed by means of a default form. Thus, in Turkish, the default “singular” form ev may mean ‘house’ or ‘houses’. By the same token, the identity feature may also take a neutral value that is outside the reference system. This does not mean that the semantics of reference break down. Just as a numerically agnostic form does not undermine the expression of plurality, a neutral value for the identity feature means no more than that the language does not have to assign a specific value in certain contexts. Thus, much of the optionality of agreement in LSE (and sign language in general) appears to stem from the matter of whether the identity feature has a specific (non-neutral) value. The question that now arises is what factors condition whether the identity feature has a specific value or the neutral value that results in a default form. As mentioned in section 6.6.2, this seems to be principally a matter of discourse rather than syntactic factors, but the matter requires 7 Equally, I am making no claims about definiteness with regard to either D or localization. Some authors have claimed that locations in space always involve definite reference (e.g. MacLaughlin 1997). Barberà (2012) provides compelling evidence that indefinite referents may be localized in space in LSC. 8 I owe the observation that point optionality and localization optionality are separate issues to Natasha Abner. Accounting for spatial agreement in LSE 287 further investigation. A loose parallel can be drawn between the use of space in sign languages and the use of prosody in a spoken language like English: various factors can play a role in shaping prosody including pragmatic, discursive and syntactic factors as well as emotional content. The use of space is also governed by multiple factors, such as metaphoric schemes (mentioned in section 3.1.1) and semantic considerations (such as the use of higher space for unspecific referents). The default situation is null marking, but if certain conditions hold, a specific value spells out as a more marked location. One such condition may be the need to contrast between different discourse referents (as we saw in section 6.5), which licenses the use of different identity values and thus of distinctive locations. Another influencing factor is iconic motivation, which provides a mapping that can account for the location assigned to present referents or topographical descriptions. The role of iconicity in the use of space is formally accommodated in the work of Schlenker (2011, 2014), and more work is needed in order to formalize other factors that trigger and influence the use of space. In summary, I claim that the identity ϕ-feature in LSE occupies a functional head that dominates the NP. There are various candidates for this position (n, Num, D; I also considered the N head itself). The parallels between identity and the referential index, on the one hand, and the association between the referential index and DP, on the other, lead me to propose that the identity feature is hosted on D, the head of DP. Once the valued identity feature is in the numeration, it may enter into agreement relations. I now turn to how such a proposal would account for spatial agreement in LSE. 7.2. Accounting for spatial agreement in LSE I have proposed that an identity feature participates in the agreement process in LSE. In this section I spell out how this process operates in terms of the Agree process stipulated within the Minimalist Program (described in section 2.3.3), by looking first at location assignment, and then at verbal agreement. I then turn to pragmatic agreement, a use of space that looks like an agreement process but that does not depend on a specific syntactic configuration. 7.2.1. Location assignment In LSE there are three mechanisms for location assignment: pointing, localization and classifier structures (see section 5.1.1). In the first case, in which a point is used to establish the location of the referent, phonological 288 Formal analysis of LSE agreement material is inserted at D, as shown in (4) for the LSE DP IXx CAT (‘the/a cat’). 9 The presence of a valued identity feature on the D head gives rise to a stem alternation at Spell-Out that generates a form with specific spatial properties (namely, the indexical point is directed towards a given locus in the signing space). Note that the phonological material inserted at D may be manual – giving rise to a manual point – or non-manual, such as eye gaze or head tilt, which may also be directed to a locus in the signing space. (4) When localization occurs, the spatial marking associated with (the specific value of) the identity ϕ-feature is expressed directly on the phonological form of the noun itself. This may be accounted for by head movement of the N to the D position so that the identity feature affixes to the lexical head N. This is shown for the LSE DP HOTELx in (5). The N head HOTEL moves up to the D head and is associated with the valued identity feature. As a result, at SpellOut the stem alternation incorporates a specific location x into the form of the noun. 9 This is somewhat similar to the analysis provided by Brunelli (2011) for location assigning indices in LIS and NGT, although here I make use of the identity ϕ-feature. Accounting for spatial agreement in LSE 289 (5) As pointed out in section 5.6 in the context of DP-internal agreement in LSE, location assignment is often predicative in nature (as evidenced by prosodic information). This is frequently the case for post-nominal points and classifier structures and I take these structures to involve some sort of reduced relative clause. In such instances the location assigning element (the post-nominal point or classifier) enters into an agreement relationship with the nominal antecedent (or, more precisely, with the valued identity feature on D). This mechanism of the agreement relationship for the identity ϕ-feature is described in the next section. 7.2.2. Verbal agreement The ϕ-feature values that appear in location assignment as described in the previous section may also enter into an agreement relationship. When this happens, an unvalued identity feature is valued by a valued identity feature by means of a probe checking a goal (as described in section 2.3.3). This may occur in the context of verbal agreement and DP-internal agreement, and I follow Carstens (2000, 2001) in assuming that Agree can equally account for the former and the latter with no need for any additional specialized mechanism. In this section, I will focus on the case of verb agreement. Furthermore, as pointed out above in section 7.1.1, I follow Danon (2011) in adopting a feature-sharing view of Agree in order to account for the fact that valued features do not appear on a single head and may be located at positions distinct from that of the lexical head with which they are associated. The DP argument of a verb bears a valued identity feature. When the verb enters into an agreement relationship with its argument, this results in 290 Formal analysis of LSE agreement the verb’s own unvalued ϕ-features becoming valued. Let us first examine the case of single argument agreement using the LSE sentence IX1 EXAMx PASSx (‘I passed the exam’). 10 In this sentence (based on example (10) in chapter 5) the verb PASS is articulated at the location associated with the referent EXAM, represented here as (6). (6) The verb’s internal argument, which occupies the complement position of V, has valued ϕ-features for identity (the value i) and number (singular). The v head contains unvalued features for number and identity and probes within its domain to find a goal with interpretable features that can value those on the probe. The DP EXAM is a suitable goal since it has valued ϕ-features and 10 Note that I continue to use the subscript 1 for signs directed towards the signer’s body, such as the indexical IX1 here. As pointed out in footnote 9 in chapter 3, this does not commit me to a first/non-first person distinction, but merely makes the form of the sign clearer. The matter will be discussed in section 8.1.2. Accounting for spatial agreement in LSE 291 there is no intervening goal between the probe and this DP. 11 Thus, the values of the goal’s features are assigned to the probe’s features, as shown by the dotted line in (6). Subsequently, the verb undergoes head movement to v, shown by the dashed line in (6), and the features affix to the lexical head, thus making it possible for the correct phonological form to be generated by the phonological stem alternation at Spell-Out, namely, the verb is articulated at the location x associated with the identity value i. 12 In the case of prototypical agreeing verbs, which show both subject and object agreement, two instances of Agree occur, such as the LSE sentence IXx xTRICK1 (‘He’s tricking me’), from example (7) in chapter 5. In the first place, the object agreement happens along similar lines to what we saw above for single argument agreement, with the unvalued features in the v head probing and receiving the values of the features on the verb-internal DP. Subsequently, when the T head is merged with the rest of the structure, the unvalued features there probe within the domain. The subject DP (in Spec-vP) serves as goal, and the features’ values are assigned to the features on T. Once again, head movement takes the verb to the T head, thus ensuring that the relevant features are affixed and available at Spell-Out so that the appropriate phonological form is generated. The subject agreement marking feature spells out as location x (associated with identity value j), and the object marking surfaces as a location on the signer’s body (associated with identity value i), yielding the form xTRICK1. The syntactic structure for this process is shown in (7). 13 The mechanism outlined here for agreement in LSE can also account for the appearance and behaviour of auxiliary verbs, such as AUX, described in section 5.3.1. When AUX appears, the Agree processes take place in the same manner as shown in (7), but the verb stays in situ in the head of V and the auxiliary is inserted into the syntactic structure directly at the head of v. The fact that there are many parallels between auxiliary verbs and light verbs, 11 I do not address the issue of whether the goal is a maximal projection or a head. Although in the text I refer to the DP, note that both features may be present on the D head. Likewise, Danon’s analysis of agreement as feature sharing allows for the features to “collect” at the head of the DP and thus be available for further agreement operations beyond the DP domain. 12 In order to account for the canonical surface form IX1 EXAM PASS, further operations are required: if the verb continues on to T, remnant movement could then displace the remaining structure higher up (to the left periphery), thus creating the order. Since my aim here is to focus on the Agree operation, I will not dwell on this matter. 13 Again, further operations are required to account for the surface form. This could be dealt with by assuming movement of the subject DP into Spec TP (to fulfil some sort of EPP constraint), or by means of remnant movement. 292 Formal analysis of LSE agreement which normally occupy v, lends support to this position for the auxiliary (see Pfau & Steinbach (2013) for the GSL auxiliary GIVE-AUX). Thus, from its initial position AUX is associated with the features valued by the object DP, and subsequently moves up to T, which affixes the features valued by the subject DP. As a result, AUX has the features required for the relevant phonological stem alternation to be applied at Spell-Out. Similarly, in the case of the PERS auxiliary, which in section 5.3.3 I argued should be considered an agreement auxiliary based on parallels with single argument agreement, insertion of the auxiliary at v would provide an adequate syntactic framework for the agreement process between PERS and the verbal argument. (7) Accounting for spatial agreement in LSE 299 means of the syntactic operation Agree, as stipulated by the Minimalist Program. In order to provide as coherent an account as possible, I propose that this syntactic agreement always occurs but may not show up in the surface form for one of two reasons. Firstly, the verb may only allow for agreement with one argument, as occurs with single argument agreement, since its phonological form does not allow the incorporation of more than one location. Secondly, the identity feature may take a default value that gives rise to a neutral location in the phonological form. This goes some way to explaining why spatial agreement looks so optional in sign languages, and also why agreement marker omission is possible. The claim here is that the syntactic process is obligatory, but the availability of a default value frequently renders the process invisible in the surface form. Finally, I distinguish between syntactic agreement proper and pragmatic agreement. In form both exploit space in a very similar manner, but syntactic agreement creates an unambiguous relationship between a verb and its arguments, whereas pragmatic agreement gives rise to an associative link that must be resolved pragmatically. I speculate that this relationship involves independent identity features with different (i.e. unmatched) values being mapped onto the same location. As such, the association is formed at the phonological level and not the syntactic level. In order to explain how a pragmatically agreeing verb obtains its own valued identity ϕ-feature, I suggest that it undergoes nominalization (by merging with n), and merges with a D head with a valued identity ϕ-feature. This allows the verb to spell out at the same location as another nominal to create a relation of pragmatic agreement between the two. This concludes the syntactic account for spatial agreement in LSE. While it does not claim to be exhaustive in addressing all the issues that such a proposal faces, the account does provide a working model that can explain the main spatial agreement phenomena found in the language, and distinguishes them from similar uses of space that do not merit a syntactic account. The framework of minimalist syntax has been applied to syntactic agreement in other sign languages (Pfau, Salzmann & Steinbach 2011 for DGS and NGT; Lourenço 2015 for Libras) and the analysis presented here provides an account for LSE and extends it to pragmatic agreement. The next section looks at a specific aspect of spatial agreement in LSE, namely a phonological constraint that appears on verbs with certain phonological characteristics, and provides a formal account for the different forms that appear in LSE. [Document text truncated for crawler view.]