Identifying, documenting, and describing marginal phonemes: A fieldworker's guide

1 Identifying, documenting, and describing marginal phonemes: A fieldworker’s guide Abstract This paper provides a guide to linguistic fieldworkers for identifying, documenting, and describing sound segments that are marginal in a language's phonological system. It explores factors that influence phonological marginality, including contrast, phonotactic and lexical constraints, variation, speaker intuitions, and frequency effects. A proposed scale helps assess a phoneme’s relative position on a continuum of marginality. Each factor’s possible impacts on a sound’s marginality are assessed. An important contribution of this study is a detailed discussion of linguistic fieldwork methods for investigating each of the factors. Special attention is given to lesser-documented languages. Keywords: marginal, phoneme, fieldwork, documentation, phonology 1. Introduction Sound segments are basic building blocks of spoken language. Of all possible speech sound types that most human beings are able to produce, each natural spoken language uses a subset. Phonological theory generally groups the sounds used in a specific variety into phonemes – abstract underlying units, which can have as surface representation one or more of the sounds that the language uses. A phoneme does not normally encode meaning by itself, but a basic characteristic of a phoneme is that it contrasts with other phonemes to distinguish meaning. An often-cited diagnostic for determining whether two sounds contrast with each other is the presence of at least one minimal pair of lexical items whose meaning depends on that contrast. For example, the English words van and fan differ by their first sound. Even if very similar – both [v] and [f] are bilabial fricatives – their distinguishing feature is that of voicing; van~fan is one of many minimal pairs distinguishing these two sounds in English. But what does a contrast actually entail? Phonemes’ function to distinguish meaning means that they perform “work” in the linguistic system. However, some phonemes may be more “productive” than others. The term “functional load” has been used as a measure of “how much ‘work’ a contrast does for creating distinctions” (Salmons 2021, 163). For example, some contrasts may be observed in many minimal pairs and in different syllable and word positions, while others may be restricted to only a few minimal pairs (or even a single one), or to only the onset or the coda of a syllable; yet others may have even more restricted phonotactic distribution, e.g. only before one specific other segment. These are examples of some criteria that can be used for determining the functional load of a phoneme. Hall (2013), and Scobbie & Stuart-Smith (2008, 106-107) list other relevant factors, and also point out that variation in functional load is variation in degree rather than in kind (Scobbie & Stuart-Smith 2008; Hall DO NOT CITE WITHOUT PERMISSION BY THE AUTHOR by Tihomir Rangelov, Preprint submitted Feb 2, 2025 2 2013).1 The main purpose of this paper is to propose fieldwork methods for investigating what factors play a role, and what role they play more precisely, in determining the marginality of sounds on a continuum. Before we turn to this discussion, some preliminaries are in order. Marginal phonemes – the topic of this volume – are sounds that are involved in contrasts within a language’s phonological system, but have a relatively low functional load. Non-marginal phonemes can also be called central (to the phonology of the language). A note is due here on the terminology around marginal phonemes. Many different terms have been used in the literature for phonemes with low functional load. Hall (2013, 216) lists a number of terms, including “semi”, “quazi”, “weak”, “partial”, “fuzzy”, “mushy”, “gradient”, or even “crazy”. In the PHOIBLE 2.0 database (Moran & McCloy 2019) the authors point out the lack of standard terminology and list as “marginal” all phonemes which may be described as, e.g. “dubious” or “occur[ing] only in loan words”. Here, in line with the approach of the entire volume, we adopt the term “marginal” (and derived terms), which also appears to be the most commonly used term in Hall’s list and the term preferred by Moran & McCloy. It is worth pointing out, however, that Hall (2013) herself prefers the term “intermediate phonological relationships” to talk about such relationships that lie “between contrast and allophony”. This choice is informed by the abovementioned fact that these relationships occur on a scale. So far, we have only discussed sounds whose relationship to other sounds can be demonstrated by minimal pairs. If no minimal pair can be found, the general principle is that these sounds are allophones (variants of a phoneme), whether their distribution is complementary or (relatively) free. However, this does not mean that all allophones are of equal standing. Some (especially complementary) allophones may be considered to have a higher functional load than others because, e.g. they are involved in near-minimal pairs, and/or because their speakers may consider them to be more salient than others. Kerswill & Williams (2002, 81) define salience as “a property of a linguistic item or feature that makes it in some way perceptually and cognitively prominent”. In other words, speakers may be more aware or less aware of a sound in their language. For example, native English speakers are not normally aware that [t] and [tʰ] both exist as possible version of the /t/ phoneme in English (Hall 2013, 223) and it can be safely assumed that the distinction between them does not play a major role in their language. Other sounds may be salient because they are found in special types of lexical items: words from a specific semantic domain, loanwords, or sound symbolic items such as onomatopoeia and ideophones. Above, we have seen that there is scalar variation in both contrasting and non-contrasting relationships (based on the at-least-one-minimal-pair diagnostic for phonological contrast). Assuming that a non-contrasting relationship is always more marginal than a contrasting one (because of the importance of the meaning-bearing function of contrast), we propose a unified 1 Moreover, languages constantly change, adapting to different pressures in their ecosystems, which adds a temporal dimension to the puzzle. 3 scale of phonological marginality (Figure 1) centred around what we call the one-minimal-pair threshold, that is, around a relationship that is based on a single minimal pair in a language. Importantly, in this paper, we will use the term “marginal phoneme” to include those salient sounds that are involved in both contrasting and non-contrasting relationships (or “regular” and “not regular” phonemes, as per Evans 2022, 1003, cf. §4). Figure 1. The marginality scale we propose as a tool for accounting for both contrastive and non-contrastive sounds in a language’s inventory. Central phonemes (which have unsubstantial distributional constraints and appear in many minimal pairs) are at the rightmost edge, while non-salient contrasts, like [t]~[tʰ] in English, appear towards the leftmost edge. The line in the middle is where a sound involved in only one minimal pair would be positioned (presuming this as a basic phonemicity and salience test, and that no other factors pull this sound away in either direction). It needs to be pointed out that the axis, the indicated spaces or cut-off points, in this figure are not meant to be categorical or proportionally represented. We have already mentioned a few factors that may determine whether a phoneme is marginal or not: the number of (near-)minimal pairs that it may be involved in, its phonotactic distribution, the speakers’ intuitions, borrowing, sound symbolism, and semantic constraints on its distribution in the lexicon. In §2, we discuss in detail these factors as well as other factors, such as pragmatic and morphosyntactic constraints, and inter-speaker variation. We also look at what can be learned from observed patterns in the phonological system, and frequency effects. The factors we discuss are well-known, they are also among those listed in Hall (2013), and Scobbie and Stuart-Smith (2008: 106-107) and are to a large extent based on previous literature. Hall already provides a detailed discussion of some of them including some theoretical implications. Last, but not least, an important issue when studying phonological marginality has to do with data availability. Considering the vast variety of factors that determine marginality, large amounts of detailed data are necessary in order to explore these relationships in detail. Of the 7000 or so languages of the world, most are undocumented or underdocumented to an extent that few facts can be inferred about their marginal phonemes. Therefore, in §3 we focus on the practical approaches to identifying, documenting, and describing marginal phonemes. This paper is thus especially aimed at linguistic fieldworkers working on lesser documented languages, especially those languages whose phonologies are not yet well studied. To our knowledge, this will be the first detailed fieldwork guide to investigating marginal phonemes. 4 2. What factors determine a sound’s marginality status? This section offers a list of factors which may be relevant to determining marginality status in phonology, and discusses each one in some detail. It is framed around the marginality of segments, but it may be applicable to other phonological phenomena. The factors listed here are based largely on existing literature, but also draw on the authors’ own experiences. Hall (2013) is an elaborate account of various reasons that may explain marginality in phonology, Scobbie & Stuart-Smith (2008, 106-107) list a relatively large number of relevant factors, while numerous other publications focus on specific factors; these are cited below where relevant. We offer short summaries and explanations; longer elaborations are made where deemed necessary. We point to existing literature where more detailed discussions can be found. The purpose of this section is to serve as a stepping stone for formulating the guidelines for the fieldworker to identify, document, and describe marginal phonemes, which follow. The factors we discuss are: lexical opposition (§2.1), lexical predictability (§2.2), borrowing (§2.3), sound symbolism (§2.4), semantic, pragmatic and morphosyntactic constraints (§2.5), phonological patterns (§2.6), inter-speaker variation (§2.7), speaker intuitions (§2.8), and frequency effects (§2.9). 2.1. Lexical opposition Labov (1994, 328) points out two factors that can help determine the functional load of a phonemic opposition: lexical opposition and lexical predictability. Lexical opposition is determined by the number of minimal pairs that depend on the distinction (assuming that such pairs exist). In other words, this is a measure of how many homonyms would be created if the two compared phonemes were to merge. The fewer those minimal pairs are, the lower the lexical opposition is, which translates as higher marginality. 2.2. Lexical predictability Labov’s (1994, 328) lexical predictability factor is a measure of how skewed the distribution of two sounds is, with higher lexical predictability indicating higher marginality. This factor has to do with the extent to which the two sounds might be phonotactically constrained, thus straddling the space between phonemicity and conditioned allophony. This is based on the fact that two sounds may not contrast in all, or most2, phonological environments. For example, in languages with final obstruent devoicing (like German), voiced and voiceless obstruents may not contrast word-finally. While this may place such sounds somewhat to the left of the rightmost edge on the marginality scale in §1, in most cases they are “still considered contrastive and thus not particularly problematic” (Hall 2013: 230). The phonotactic restrictions may be various, e.g. a consonant may be restricted to either the syllable onset or coda, or to a certain position in a 2 In fact, arguably, it is somewhat rare that two sounds contrast in all possible environments. 5 consonant cluster; a sound may only appear in specific syllables in the word (e.g. rounded vowels may only occur in the first syllables of words in some Uralic languages, (see chapter** in this volume?); a consonant may be restricted to before a certain type of vowels (e.g. linguolabials in some Oceanic languages appear only before non-front vowels, see Lynch 2020) or before a specific vowel (e.g. bilabial trills tend to appear only before /u/ in most languages, see Maddieson 1989). Arguably, these examples can be placed on a scale from less substantial to more substantial constraints, respectively. 2.3. Borrowing The relationship between marginal phonemes and loanwords is well known (cf. intro to this volume). Language contact provides an opportunity for sounds from one language to enter the system of another through lexical borrowings. While lexical borrowings are often adjusted to the receiving language’s phonotactics, borrowed words frequently retain some features of their original pronunciation in the donor language. When a sound that is not native to the recipient language is retained in borrowings, such a sound can be termed a xenophone, and is expected to be more marginal than the recipient language’s central phonemes because it is likely used in relatively few lexical items. From a statistical point of view, it is less likely for xenophones to have high lexical opposition and low lexical predictability because of their restricted distribution in the lexicon. Some examples of xenophones are: French borrowings into German may retain nasal vowels, not otherwise found in German, like the final vowel in Restaurant for many German speakers; the glottal and uvular plosives in Persian appear in Arabic borrowings (Bijankhan 2018); voiced plosives in Finnish appear (mostly) in words borrowed from various languages (Suomi et al. 2008); in Chamic languages (of the Austronesian family), implosives are found almost exclusively in loanwords from Mon-Khmer languages (of the Austroasiatic family). Retaining the donor language’s sounds, rather than adjusting the pronunciation, may largely depend on the receiving language’s speakers’ fluency in the donor language, that is, on patterns of multilingualism and speaker attitudes. In some cases, loanwords may not be introducing a new sound to a language, but may contribute to a native allophone’s “promotion” towards a more phonemic (i.e. less marginal) status. Hall (2013: 238) provides an example from English where [s] and [ʃ] do not normally contrast in initial consonant clusters in native vocabulary, but Yiddish loanwords such as schlep [ʃlɛp], schmooze [ʃmuz], spiel [ʃpiɫ], and schtick [ʃtɪk] can form (near-)minimal pairs with native words. It should be pointed out that there might be variation in how and to what extent loanwords are adjusted in the recipient language, and to what extent and under what conditions borrowed sounds spread through its lexicon. An example of a detailed study on these matters is Bostoen 6 and Donzo (2013) on the labial-velar stops in Lingombe, a Bantu language spoken in DR Congo, which were borrowed from neighbouring non-Bantu languages and later spread through the lexicon of Lingombe due to sound symbolism and speaker attitudes. 2.4. Sound symbolism Some marginal phonemes may be linked to sound symbolism. Sound symbolism is observed in lexical items, in which there is a perceptual connection between the phonetic form and the concept being expressed. It can take different shapes, the most common ones being onomatopoeia and ideophones. In onomatopoeia, the sound of the word imitates its meaning, like the English word purr to denote a sound made by cats. An ideophone is “a member of an open lexical class of marked words that depict sensory imagery” (Dingemanse 2019). Ideophones are particularly common in some African, East Asian and Southeast Asian languages. The unusual phonetic structure involved in sound symbolic items means that they often contain phones that are otherwise not used in the language, but have a “special dramaturgic function” (Voelz & Kilian Hatz 2001, 3). Sometimes, sounds employed in sound symbolic items are crosslinguistically rare sounds such as clicks or bilabial trills. It may be the case that such sounds share little phonetic similarity with any other sounds in a language’s phonology, which may make it difficult to identify sounds to compare them with (cf. Footnote 4 in §2.6). This alone may be a reason to consider such sounds as salient, albeit marginal, parts of the language’s phonological system (Hall 2013, 224). Cross-linguistically common sounds can also perform sound symbolic functions (see e.g. Erben Johansson et al. 2020; Ibarretxe-Antuñano 2019). Thus, sound symbolism may also provide a mechanism for non-native sounds to enter the phonological system of a language and even become contrastive. Such is the case with /p/ and /ŋ/ in the Cariban language Katuena (Smoll 2015), palatalised stops in Pastaza Quichua (Nuckolls et al. 2016), or /ɸ/ in Japanese (Hamano 2019). 2.5. Semantic, pragmatic, and morphosyntactic constraints Another factor that may play a role in the marginality of a sound is whether it occurs in contexts that are semantically, pragmatically and/or morphosyntactically restricted. Some examples follow. Evans (2022) describes three cases: In the language Nen, spoken in Papua New Guinea, the sounds [ã,ẽ,h] are restricted to interactional contexts (interjections and deictics), which demonstrates aspects of all three types of constraints discussed in this subsection. In English, [θ,ð] are only contrastive word-finally, and then mostly in a paradigm where the words with [θ] and [ð] form noun-verb pairs, respectively, like wreath [θ] (n.) ~ wreath [ð] (v.); teeth [θ] (n.) ~ teeth [ð] (v.); mouth [θ] (n.) ~ mouth [ð] (v.), i.e. the contrast is restricted to these word class paradigms. On the other hand, word-initially there are no [θ]~[ð] contrasts, and, more importantly to the point here, [ð] is found only in words such as thou, that, then, the, though, which are morphosyntactically restricted to closed-class function words, and semantically 7 restricted to meanings of person, space, time, and discourse deixis. Evans’s third example comes from the Australian language Bininj Kunwok, where the glottal stop is primarily used in a morpheme (which comprises the glottal stop on its own) with three different grammatical functions. Finnish offers another example of a morphosyntactic constraint: in Finnish native vocabulary, [d] and [ŋ] appear only in the weak grade of morphemes with consonant gradation (see chapter *** in this volume), i.e. their distribution in native vocabulary is dependent on a morphophonological condition, which arguably increases their marginality status. On the other hand, the Oceanic language Avava, spoken in Vanuatu, offers an example of morphophonological conditioning, which can be argued to decrease the marginality of a sound. Namely, the marginal prenasalised bilabial trill phoneme in this language, which normally occurs only before /u/, can also surface before other vowels as a result of the following morphophonological process: where the third person singular irrealis prefix /ᵐbʷV- ~ ᵐbV-/ appears before a verb root beginning with /v-/, /w-/ or /vʷ-/, which is followed by a vowel other than /u/, the prefix vowel is elided and a [ᵐʙ] systematically surfaces, as in: /ᵐbʷe-vʷel/ [ᵐʙel] ‘3SG:IRR-come’, /ᵐbi-vi/ [ᵐʙi] ‘3SG:IRR-COP (he/she/it will be)’ (Crowley 2006, 31–32). As demonstrated by the examples, this process is observed in core and frequent vocabulary, which suggests lower marginality (cf. §2.9). Avava’s closely related neighbouring language Neverver, exhibits a similar process – when the irrealis prefix mis attached to verb stems beginning with [β], the fricative is clearly trilled as in /im-βu/ [iᵐʙu] ‘3SG:IRR-go’ (Barbour 2012: 34, 174). In Neverver, this process may contribute to higher type and token frequency but the effect on decreasing marginality is lower than in Avava, because it does not produce bilabial trills before vowels other than /u/ (in Neverver, an [u] is inserted after the trill if the following vowel is not [u]) and thus does not result in lower lexical predictability. Another important factor involves the morphosyntactic properties of the items in the minimal pairs. For example, if a minimal pair involves items of the same word class (and even subclass), and with the same grammatical marking, the lexical opposition may be more likely to persist because it does some heavy lifting that otherwise morphology, syntax and pragmatics would do. Or, as Van de Veen & Medjo Mvé (2010, 9) put it: “because of their ability to commute in identical or similar grammatical contexts, to potentially belong to the same paradigm”. An example comes from Icelandic voiceless sonorants, such as /r „/, which are already marginal phonemes since they only contrast with their voiced counterparts word-initially (Árnason 2005). Importantly, voiced and voiceless sonorants in Icelandic mostly form minimal pairs where the two members either cannot belong to the same paradigm (because they are from different word classes) or mark different morphosyntactic categories, making them pragmatically already well distinguishable. This suggests a relatively low workload of such phonemic contrasts. Examples of such minimal pairs in Icelandic are reyna [reɪna] ‘to try (verb, infinitive)’ ~ hreina [r „eɪna] 8 ‘clean (adj., weak declension)’, and rúta [ruːta] ‘bus (feminine, NOM.SG)’ ~ hrúta [r „uːta] ‘ram (masculine, NOM/GEN.PL)’.3 These are just a few examples of the semantic, pragmatic, and morphosyntactic constraints that can determine sounds’ marginality status. As they demonstrate, these constraints can operate within different domains of language structure, and the range of possible constraints is vast. In the case of lesser-documented languages, these will be uncovered as the documentation process progresses, demonstrating the iterative nature of work related to marginal phonemes (see §3). 2.6. Patterns in the phonological system The place of a sound in the phonological system of a language is also a factor in determining its marginality. One important question is whether the sound groups with other sounds in the system according to shared features (e.g. a plain voiceless plosive in a language with a row of other voiceless plosives already established as central phonemes).4 In other words, one needs to take into account how many features a sound shares with other sounds and what features tend to be contrastive in this language. Generally, it can be presumed that a sound that fits well into the phonological system of the language is less likely to be marginal. This has to do with the prominent question in phonology about the tendency for symmetry in phonological systems and its role in language evolution (see Blevins 2004, 281-285 for a detailed discussion of this elaborate issue). A marginal phoneme may occupy a place that would have otherwise constituted a gap in a phonological system. Bye (2013, 41) offers the example of English /ʒ/ which is a marginal phoneme found mostly in French loanwords, but also filled a gap in the existing fricative inventory of English, which had a voiced-voiceless distinction in its fricative series, except at the postalveolar place where only /ʃ/ existed. In another example, Rangelov et al. (2023) argued that the emergence and persistence of the very typologically rare plain voiceless bilabial trill [ʙ „] in some languages of Malekula Island in Vanuatu, was supported by an already existing gap in the trill inventory of these languages where there was already a prenasalization contrast in alveolar trills and a prenasalised bilabial trill. In some of these languages, the plain trill is contrastive and thus a (marginal) phoneme, while in others it appears to not be contrastive, but highly salient among speakers (cf. §§2.8, 3.6). Marginal phonemes may also be leftovers of previously central phonemes that underwent a change. Thus, in Ahamb, one of the Malekula languages with bilabial trills, Proto-Oceanic 3 In fact, many of these minimal pairs exist because of Icelandic’s rich morphology. 4 A related detail here is that for a sound to be considered for a contrasting or allophonic relationship, it is generally agreed that it needs to share some phonetic similarity – i.e. to share at least some phonological features – with the sounds it is compared or grouped with (Pike 1947). It is possible that sounds that are outliers in the language’s phoneme inventory (i.e. they share few features with other sounds in the system, e.g. a single implosive or click sound in a language with no other such sounds), may persist by means of high salience, e.g., because they are employed in loanwords or sound symbolic items. By definition, an outlier sound should be considered marginal, unless other factors suggest otherwise. 9 prenasalised alveolar plosives likely changed to prenasalised alveolar trills (Tryon 1976), leaving a gap in the plosive inventory that involved a prenasalization contrast at the labial, alveolar, and velar places of articulation. The only two words with [ⁿd] that have been documented in contemporary Ahamb are [ⁿdasⁿdas] ‘smooth’ and [naliⁿdumⁿdum] ‘whale’ (Rangelov 2020). In these examples, sound symbolism may play a role in disfavouring trilling, as trilling is known to be associated with roughness (Winter et al. 2022), and [ⁿdumⁿdum] arguably better represents the size of a whale and the noises associated with its movements than *[ⁿrumⁿrum]. It is important to note here that it may be relevant to determine the marginality not just of a specific sound but of a sound type within a phonological system. For example, Güldemann (2007), Sands & Gunnink (2019), and Breniznger & Shah (2023) explore the marginality of clicks in different African languages for the purpose of comparing whether clicks are more or less marginal in different groups of languages, rather than within a single language (see §2.9 for more details on these studies). Lastly, the size of the (hypothesized) phoneme inventory for a language may matter for some of the factors described in this section (Dockum & Bowern 2019, 49). A larger phoneme inventory means that, statistically, there will be fewer minimal pairs per contrast on average, since the lexicon is finite, or that an observed skewed phonotactic, semantic, pragmatic or morphosyntactic distribution may be due to chance than actual phonotactic constraints (especially when the amount of data is limited). This problem is more apparent when dealing with extreme cases of languages with over a hundred phonemes, but it should also be considered in more “trivial” cases. What is important, however, is that in a larger-sized inventory more phonemes may be more likely to be marginal, which makes the averages mentioned above become less meaningful and the variation in the phonemes’ distribution and other factors all the more important. Bulgarian palatal consonants offer a case in point. In traditional accounts of Bulgarian phonology (e.g. Stoykov 1951; Tilkov 1982), there are 16 palatal consonants (out of a total consonant inventory of 37), which are considered phonemic, albeit with limited distribution. They only appear before non-front vowels, unlike in other related languages like Serbo-Croatian and Russian, where they are found before all vowels and in the syllable coda. Whether palatal consonants in Bulgarian should be considered (marginal) phonemes or not is beyond the scope of this paper, but the importance of understanding factors like their distribution and phonetics (e.g. Simeonov 1984; Pritchard 2012) becomes clear. Namely, once we understand their behaviour, the “working size” of Bulgarian’s consonant inventory decreases substantially to very average levels (cf. Maddieson 2013). 2.7. Inter-speaker variation A sound may be produced in various ways by different speakers of a given variety. Hall (2013, 239) uses the term “variability” of production to refer to differences between speakers where 16 acoustic and perceptual studies are discussed briefly at the end with references for further reading. When it comes to the documentation and analysis of articulatory data, one problem with many methods is that some of the necessary equipment is very bulky, expensive and/or requires infrastructure or conditions that may be impossible to guarantee in every field setting. If the researcher has access to such equipment, bringing speakers to a phonetic lab may or may not be possible or practical. Here, we want to bring attention to four relatively easy, (mostly) noninvasive techniques for documenting and analysing some articulatory features in virtually any field site with relatively little preparation, expertise or equipment requirements. Some of the techniques we present here have become a lot more accessible in the past couple of decades also due to technological developments. We discuss slow-motion video recordings, linguography/palatography, the Earbud Method for investigating nasality, and the use of portable ultrasound devices. For any labial sounds or sound features, a slow-motion recording can help track the movement of the lips (or tongue, if it is visible, e.g., in linguolabial sounds). Painting dots (e.g. with lipstick) on points of interest (e.g. middle of lips, edges of mouth) is an easy way to later track these points with precision in computer applications. Using this method, labial or labialised sounds can be studied (e.g., Rangelov 2019). Very portable cameras with capabilities of recording 240 frames per second or higher frame rates are widely available for a reasonable price. Such recordings may also be useful for tracking tongue movement or other features in phonetic environments where the mouth remains open and they may be visible. Linguography and palatography are well-established methods for determining the nature of the contact between the tongue and the hard palate or back of the teeth. This can provide detailed information about the exact place and manner of articulation of a large number of coronal consonants. This method is easy to learn and very low-cost, requiring readily available materials. Linguography involves painting the palate with a mixture of medical charcoal and olive oil using a brush, then asking the speaker to pronounce a target token designed to isolate the single coronal contact, then asking the speaker to protrude their tongue, and photographing it. Palatography involves painting the protruded tongue in the same way and, later, the insertion of the mirror into the speaker’s wide-open mouth in order to record an image of the contact area on the palate. This means that this method is somewhat invasive, albeit with very small risks involved. Hygiene measures need to be extra strictly observed, and speakers need to be informed about the inconveniences involved before they give their informed consent.8 Ladefoged (2003, 36ff), Anderson (2008), and Butcher (2013) and provide more detail on these techniques. 8 Informed consent, and other ethical considerations, are an important part of any fieldwork activity. Standard informed consent and ethical considerations need to be taken into account when working with native speakers of a language on any topic, including phonology. Discussing ethical issues regarding fieldwork is beyond the scope of this paper, but all best practices should be followed at all times. Most of the data collection methods in this section 17 A relatively easy and very low-cost method for analysing nasality in virtually any field setting is the Earbuds Method first described by Stewart & Kohlberger (2017) and refined by Bruil & Stewart (2022). This method involves plugging a set of cheap earbud headphones into the microphone jack of a recorder and making a stereo recording of the speaker pronouncing targets while holding one earbud very close to their mouth and the other earbud at the entrance of one of their nostrils. This way, the relative intensity of the airflow coming out of the mouth and the nose is captured in two different channels with little interference. These recordings can then be analysed in Praat (Boersma & Weenink 2025) and allow for identification of the onset and offset (and duration) of nasal gestures and closures of the oral cavity. The details of how to perform recordings using this method, its limitations, and how to analyse the recorded data, are described in detail in the two works cited above. This method has been used to investigate various phenomena such as prenasalisation (e.g., Rangelov 2019; 2023; Seara et al. 2021), voiceless nasals (e.g., Lovick et al. 2023), and in sociophonetic studies (e.g., Crosby & Dalola 2023 on a Korean baby talk register). Carignan (2024) offers a validation of the method, which confirms that it is reliable for establishing general facts about nasality, but more accurate methods may be necessary for establishing fine detail, which may or may not be necessary for the purposes of exploring marginal phonemes. Ultrasound tongue imaging can be used in virtually any field setting more easily since the development of portable ultrasound systems. Such portable systems are not as affordable as the other methods described above, and they may require more resources in terms of power supply (also possible through a laptop, for example), but they offer a unique opportunity to track the position of the entire upper surface of the tongue at different points in time, in sync with an audio input. This can provide articulatory data that can shed light on the features of some sounds that may be key for determining their phonemic status. Some systems that are designed especially for phonetic research provide additional functionalities such as synced video tracking of the lips (e.g., Articulate Instruments 2024), which can be useful for exploring some complex segments and labiovelarisation, for example. Exploring the acoustic properties of sounds usually involves recordings with widely available professional microphones that are part of the standard toolkit of fieldworkers. Even recordings made outside of a soundproofed space can be useful as long as noise levels are not excessive. For analysis, there are standard tools like Praat (Boersma & Weenink 2025). Finally, it may also be relevant to design perceptual experiments in order to explore auditory properties of the studied sounds. This can be done by playing unaltered or edited recordings to fall under general data collection methods and the same ethical considerations apply. The reader is encouraged to consult the latest literature on the topic and follow all institutional requirements and guidelines and the speaker’s wishes. 18 speakers and asking for judgements. This is a form of elicitation using audio stimuli and was discussed briefly in §3.1. More details and examples can be found in, e.g., Maddieson (2002, 425-426). 3.5. Sociolinguistic studies The methods of sociolinguistics can be used to inform our understanding of various factors outlined in §2 when studying specific cases of marginality. The methods of the subfield of sociophonetics (Kendall & Fridland 2021; Docherty 2022) can be used to study phonetic variation, which can inform frequency effects and lexical opposition/predictability. These methods are closely related to methods outlined above on phonetic research (§3.4) and elicitation (§3.1). More general sociolinguistic methods, including sociolinguistic interviews and participant observation (Meyerhoff et al. 2011; Schilling 2013) can shed light on these factors as well as speaker’s intuitions and attitudes. These methods may require extensive preparation by the fieldworker. 3.6. Orthography development It is generally advised that when working on a lesser-documented language, especially one whose phonology has not yet been analysed in detail, the fieldworker should initially use phonetic transcription. A working orthography can complement and even replace phonetic annotation once the phonological system has been established with some confidence. Very often, language documentation projects also aim to provide an orthography for the community as part of language revitalization/maintenance/conservation efforts. There are many intricacies and pitfalls associated with designing a community orthography; these are beyond the scope of this paper and the reader is encouraged to review the extensive literature on the topic (e.g., Jones & Mooney 2017). In all cases, community orthographies should involve the active participation of the speaker community and here we focus on how discussions between the community and the fieldworker, and observations of how community members use the orthography, can inform their understanding of phonological marginality in their language. Discussions regarding a community orthography and observations of speakers using the orthography can be a source of information about the speakers’ intuitions and perceptions of different sounds and the phonological system. They may reveal what sounds or contrasts (different) speakers may or may not find to be salient, thus informing both lexical opposition/predictability hypotheses and variation hypotheses. Very importantly, they may also reveal speakers’ intuitions. For example, community members may prefer to adopt a dedicated grapheme for a marginal sound, because they consider it to be salient enough. There are different reasons why speakers may find a sound to be more salient than a purely phonological analysis may suggest. In §2.8 (including Footnote 5), we discussed identity attachment to bilabial trills and linguolabials in some Oceanic languages. In the case of the Lamap language discussed there, 19 one of Williams’s (2019) reason to treat the plain trill as a (marginal) phoneme is that speakers told her that they preferred to have a separate notation for it in the practical orthography that was being designed for this language. They proposed the digraph <pp>, by analogy with the established <bb> digraph for the prenasalised voiced bilabial trills, which demonstrates the analogy they perceive between the two sounds. Additionally, it may also be useful for a fieldworker to observe how speakers prefer to spell borrowings and sound symbolic items, which may shed light on related phonetic and phonological relationships. This method presents some pitfalls, however. Firstly, speakers may intuitively prefer phonetic rather than phonemic orthography, that is, they may spell conditioned allophones differently, or exhibit variation in spelling free allophones. Secondly, the speakers’ literacy skills need to be taken into account, and more specifically, the speakers’ mastery of orthographies for other languages, which may heavily influence their choices. These may be non-standard orthographies of other minority languages or (often standardised) orthographies of more prestigious languages that are taught in school or used by religious institutions, such as lingua francas; related prescriptive attitudes may play a role too in speakers’ orthography choices and use. 3.7. Observations of L1 acquisition It is well known from research on first language (L1) acquisition that newborns are able to discriminate most sound contrasts that exist in natural languages, and they lose this ability to focus on discriminating the sound contrasts in the language(s) to which they are exposed (Werker & Tees 1984). The mechanisms that underpin this “learning by forgetting” (Mehler & Dupoux 1994) are based on infants being sensitive to statistical distributions in the input, i.e., they are able to create phonemic categories by tracking the frequency of segments in the input (Gervain & Mehler 2010, 201). From this follows that infrequent sounds in the language, such as marginal phonemes, may be acquired later than other phonemes in the language. The fieldworker might then observe that a marginal contrast may be absent in very young children and this can be used to form a hypothesis regarding marginality. These observations may be made by the fieldworker themselves or by asking caregivers about their observations and impressions. For ethical purposes, we advise fieldworkers to be open about such activities, inform caregivers and their consent to report such observations even if they are made unintentionally during routine daily activities. If possible, or deemed necessary, recordings of children can be made too. Graham et al. (2013) present extensive general guidelines for ethical research involving children. 20 4. Discussion and conclusions In this paper, we first showed that the sounds used in a language can occur along a marginality scale, and we discussed various factors that can determine the marginality of a sound (whether it is considered phonemic or not). The factors we discussed in detail in §2 had to do with distribution in the lexicon, phonotactic constraints, patterns in the broader phonology of the language, variation, speaker intuition, and type and token frequency. Based on this, in §3, we offered a number of methods that the linguistic fieldworker can use in order to identify, document, and describe marginal phonemes, especially in understudied languages. In this section, we offer a summary of (1) the role of the different factors on the positioning of a phoneme along the marginality scale, and (2) the relationships between the proposed methods and the factors. Then we offer a few concluding remarks. We saw that the factors discussed in §2 can predict a phoneme’s marginality, that is, different factors can position it in a particular section of the marginality continuum or “pull” it to the left or right of the marginality scale. The proposed relative effects of different factors are illustrated in Figure 2. It is one of the researcher’s tasks to investigate and hypothesize how strong a pulling force is. It is important that the effects illustrated in Figure 2 are based on each factor operating alone. In reality, as we saw in a number of examples above, these factors normally operate in combination. Besides the assumption that the one-minimal-pair threshold plays a significant role, all other scales and cut-off points in Figure 2 are relative. In this paper, we have been careful not to imply any quantitative evaluations, because this is beyond its scope. However, we believe that this model can be a stepping stone towards a quantitative model for determining phonological marginality. This would, however, require more theoretical input and, very importantly, a lot more detailed and systematically organized data on many phylogenetically and geographically diverse languages. As pointed out already, such data are particularly missing for many lesser studied languages, including for entire large language families, and for languages that operate in diverse language ecosystems. In the meantime, as already discussed in §2.9, type and token frequency can be used for quantitative analyses as a proxy for the sum of the effects of the different factors in combination, as illustrated by Güldemann (2007), Dockum & Bowern (2019), Sands & Gunnink (2019), and Breniznger & Shah (2023). 21 Figure 2. The effects that factors discussed in §2 can have on a sound’s position on the marginality scale. The representations are meant to be illustrative, not proportional. We now turn to a summary of what fieldwork methods can be (directly) used to explore the different factors listed in §2, based on the discussions in §3. In Table 1, we see that a lot can be achieved by employing many of the traditional methods of language documentation, which the experienced fieldworker should be well versed in – elicitation, corpus exploration, data organization. Orthography development work can also help form hypotheses regarding many of the factors, even if this method comes with many caveats (§3.6); this is another activity that many linguistic fieldworkers engage with already. Thus, it is clear that for a fieldworker to be in a better position to explore marginal phonemes, the amount of necessary additional resources may be relatively small. Namely, some fieldworkers may need to add to their expertise on marginal phonemes (by using the information in this paper, among other sources), and to prepare in advance – ideally at the outset of a language documentation project – to look out for relevant data. The advice on data organization in §3.3 alone can go a long way. Two of the remaining methods – phonetic and sociolinguistic studies – may require more substantial additional resources in terms of preparation, technology and/or skills. However, the returns from such studies can be substantial so the fieldworker is encouraged to employ such methods when the resources allow. Finally, observations regarding language acquisition by small children can provide specialized frequency data, which can complement the other methods in forming and testing hypotheses regarding marginality. 22 Apply this method → to study this factor ↓ Elicitation (§ 3.1) Corpus exploration (§ 3.2) Data organization (§ 3.3) Phonetic studies (§ 3.4) Sociolinguistic Studies (§ 3.5) Orthography development (§ 3.6) L1 acquisition observation (§ 3.7) Lexical opposition (§2.1) ✓ ✓ ✓ ✓ ✓ ✓ Lexical predictability (§2.2) ✓ ✓ ✓ ✓ ✓ ✓ Borrowing (§2.3) ✓ ✓ ✓ ✓ ✓ ✓ Sound symbolism (§2.4) ✓ ✓ ✓ ✓ ✓ Sem., pragm., morphosynt. constraints (§2.5) ✓ ✓ ✓ Patterns in phonology (§2.6) ✓ ✓ ✓ Inter-speaker variation (§2.7) ✓ ✓ ✓ ✓ ✓ ✓ Speaker intuitions (§2.8) ✓ ✓ ✓ Frequency (§2.9) ✓ ✓ ✓ ✓ Table 1. A summary of the applicability of the methods discussed in §3, to the investigation of factors discussed in §2. Needless to say, the methods discussed in §3 should ideally be used in combination. As we already saw, the process of analysing marginal phonemes is iterative and the findings of one method can provide data for applying another method, which can feed other methods or back to methods already used. We demonstrated this with a few examples. Explorations of bilabial trills of languages spoken in Vanuatu, discussed in a few subsections, showed that their marginality status in individual languages can be determined by exploring their phonotactics, lexical constraints, inter-speaker variation, speakers’ intuitions, frequency, and their place in the phonological system, by using virtually all the proposed methods as possible and deemed necessary. Similarly, in his study of Nen’s marginal sounds [ã,ẽ,h] and their restricted semantic/pragmatic distribution (§2.5), Evans (2022, 999-1003) complements that evidence with explorations of Nen’s phonological system, frequency effects, borrowing, sound symbolism, lexical opposition and predictability, and inter-speaker variation to try to understand the marginality status of these sounds. Evans concludes that “[t]he most convincing answer is 23 probably to say that they are phonemes, but not regular ones: that the phonology is structured around a core of fully regular phonemes augmented by an outer layer of more restricted ones”. This conclusion summarises well the theoretical debates around phonologically marginal phonemes. For the fieldworker and documenter, perhaps the more important task at hand is to try to describe such phonemes in as much detail as possible in order to try to determine their position along the marginality scale. Hopefully, this guide provides a good starting point for such noble efforts. References Anderson, Victoria B. 2008. Static palatography for language fieldwork. Language Documentation & Conservation 2(1):1–27. Árnason, Kristján. 2005. Íslensk tunga. Handbók um hljóðfræði og hljóðkerfisfræði / Bd 1, Hljóð. Reykjavík: Almenna bókafélagið Articulate Instruments. 2024. Ultrasound for phonetic research. https://www.articulateinstruments.com/ultrasound-for-phonetic-research/ (Accessed on February 2, 2025) Barbour, Julie. 2012. A grammar of Neverver. Berlin: De Gruyter Mouton. Bijankhan, Mahmood. 2018. Phonology. In Anousha Sedighi & Pouneh Shabani-Jadidi (eds.), The Oxford Handbook of Persian Linguistics. Oxford University Press. Blevins, Juliette. 2004. Evolutionary phonology: The emergence of sound patterns. Cambridge:Cambridge University Press. Boersma, Paul and David Weenink. 2025. Praat: doing phonetics by computer [Computer program]. Version 6.4.27, retrieved 1 February 2025 from http://www.praat.org/ Bostoen, Koen and Jean-Pierre Donzo. 2013. Bantu-Ubangi language contact and the origin of labial-velar stops in Lingombe (Bantu, C41, DRC). Diachronica 3010.1075/dia.30.4.01bos Bowern, Claire. 2008. Linguistic fieldwork: A practical guide. Basingstoke, New York: Palgrave MacMillan. Brenzinger, Matthias & Sheena Shah. 2023. A typology of the use of clicks. Stellenbosch Papers in Linguistics Plus 67. 59-77 Bruil, Martine & Jesse Stewart. 2022. Phonetics and phonology of nasality in Ecuadorian Siona. Phonological Data and Analysis 4(3). 1-34. https://doi.org/10.3765/pda.v4art3.44 Butcher, Andrew. 2013. Research methods in phonetic fieldwork. In M. Jones & R.-A. Knight (Eds.), The Bloomsbury companion to phonetics. 57– 78. New York, NY: Bloomsbury Academic. Bye, Patrik. 2009. Three types of marginal contrast. Paper presented at the Torontø-Tromsø Phonoløgy Workshøp, Toronto, ON, October 9–11. Bye, Patrik. 2013. The lexicon has its grammar, which the grammar knows nothing of: Marginal contrast andphonological theory. Nordlyd 40(1). 41-54. Special Issue: A Festschrift on the occasion of X years of CASTL phonology and Curt Rice's Lth birthday, edited by Sylvia Blaho, Martin Krämer, and Bruce Morén-Duolljá Carignan, Christopher. 2024. Ground-truth validation of the “earbuds method” for measuring acoustic nasalance. J. Acoust. Soc. Am. 156 (2): 851–864. https://doi.org/10.1121/10.0028122 24 Crosby, Drew and Amanda Dalola. 2023. Quantifying nasalance in Korean aegyo. In: R. Skarnitzl & J. Volín (Eds.) Proceedings of the 20th International Congress of Phonetic Sciences – ICPhS 2023. 3597-3601. International Phonetic Association. Crowley, Terry. 2007. Field Linguistics: A Beginner’s Guide. Edited and prepared for publication by Nick Thieberger. Oxford: Oxford University Press. Dimock, Laura. 2009. A grammar of Nahavaq (Malakula, Vanuatu). Victoria University of Wellington. (PhD thesis.) Dingemanse, Mark. 2019. ‘Ideophone’ as a comparative concept. In Kimi Akita & Prashant Pardeshi (eds.), Ideophones, Mimetics and Expressives, 13-34. John Benjamins. Docherty, Gerard. 2022. Sociophonetics. In Oxford Research Encyclopedia of Linguistics Dockum, Rikker and Claire Bowern. 2019. Swadesh lists are not long enough. Language Documentation and Description 16. 35-54. ELAN (Version 6.9) [Computer software]. 2024. Nijmegen: Max Planck Institute for Psycholinguistics, The Language Archive. Retrieved from https://archive.mpi.nl/tla/elan Erben Johansson, Niklas, Andrey Anikin, Gerd Carling and Arthur Holmer. 2020. The typology of sound symbolism: Defining macro-concepts via their semantic and phonetic features. Linguistic Typology 24(2). 253-310. https://doi.org/10.1515/lingty-2020-2034 Evans, Nicholas. 2022. Pushing the boundaries: Marginal phonemes and dialogic interaction. Russian Journal of Linguistics 26(4). 995-1011. Fieldworks [computer software]. 2024. Dallas: SIL International. (Version 9.1.25. Retrieved from https://software.sil.org/fieldworks/download/) Gervain Judit and Jacques Mehler. 2010. Speech perception and language acquisition in the first year of life. Annu Rev Psychol. 61. 191-218. Graham, A., Powell, M.A., Taylor, N., Anderson, D. & Fitzgerald, R. 2013. Ethical Research Involving Children. UNICEF: Florence Güldemann, Tom. 2007. Clicks, genetics, and “proto-world” from a linguistic perspective. University of Leipzig Papers on Africa 29. Hall, Kathleen Currie. 2013. A typology of intermediate phonological relationships. The Linguistic Review 30(2). 215-275. https://doi.org/10.1515/tlr-2013-0008 Hamano, Shoko. 2019. Monosyllabic and disyllabic roots in the diachronic development of Japanese mimetics. In Kimi Akita & Prashant Pardeshi (Eds.), Ideophones, Mimetics and Expressives, 57-77. John Benjamins. Ibarretxe-Antuñano, Iraide. 2019. Towards a semantic typological classification of motion ideophones. In Kimi Akita & Prashant Pardeshi (Eds.), Ideophones, Mimetics and Expressives, 137–166. John Benjamins. Jones, Mari C., and Damien Mooney, eds. 2017. Creating orthographies for endangered languages. Cambridge University Press. Kendall, Tyler and Valerie Fridland. 2021. Sociophonetics. Cambridge University Press. Kerswill, Paul & Ann Williams. 2002. ‘Salience’ as an explanatory factor in language change: Evidence from dialect levelling in urban England. In Mari C. Jones & Edith Esch (eds.), Language change: The interplay of internal, external and extra-linguistic factors, 81–110. Berlin: Mouton De Gruyter. Labov, William. 1994. Principles of linguistic change. Volume 1: Internal factors. Oxford: Blackwell. Ladefoged, Peter. 2003. Phonetic data analysis: An introduction to fieldwork and instrumental techniques. Oxford: Blackwell. 25 Lovick, Olga, Kristy Reyes Herrera and Jesse Stewart. The realization of voiceless nasal consonants in upper Tanana (Dene). In: R. Skarnitzl & J. Volín (Eds.) Proceedings of the 20th International Congress of Phonetic Sciences – ICPhS 2023. 3281-3285. International Phonetic Association. Lynch, John. 2020. The Bilabial-to-Linguolabial Shift in Southern Oceanic: A Subgrouping Diagnostic? Oceanic Linguistics 58(2). 292-323 Maddieson, Ian. 1989. Aerodynamic constraints on sound change: The case of bilabial trills. UCLA Working Papers in Phonetics 72. 91-115 Maddieson, Ian. 2002. Phonetics in the Field. Annual Meeting of the Berkeley Linguistics Society 28. Maddieson, Ian. 2013. Consonant Inventories. In: Dryer, Matthew S. & Haspelmath, Martin (eds.) WALS Online (v2020.4) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.13950591 (Available online at http://wals.info/chapter/1 , Accessed on 2025-02-07.) McKerras, Ross. 2001. Uripiv phonology and grammar. SIL International. Meakins, Felicity, Jennifer Green & Myfany Turpin. 2018. Understanding Linguistic Fieldwork. London: Routledge. Mehler, Jacque and Emannuel Dupoux. 1994. What infants know: The new cognitive science of early development. Cambridge, MA: Blackwell Meyerhoff, Miriam, Chie Adachi, Golnaz Nanbakhsh and Anna Strycharz. 2011. Sociolinguistic Fieldwork. In Nicholas Thieberger (ed.), The Oxford Handbook of Linguistic Fieldwork. Oxford University Press. Moore, Alice. 2019. A grammar sketch of Uripiv. Hamilton, New Zealand: University of Waikato. (MA thesis.) Moran, Steven and McCloy, Daniel (eds.) 2019. PHOIBLE 2.0. Jena: Max Planck Institute for the Science of Human History. (Available online at http://phoible.org, Accessed on 2023-1127.) Nuckolls, Janis B., Elizabeth Nielsen, Joseph A. Stanley and Roseanna Hopper. 2016. The systematic stretching and contracting of ideophonic phonology in Pastaza Quichua. International Journal of American Linguistics 82(1). 95-116. https://www.jstor.org/stable/26559045^doi: https://www.jstor.org/stable/26559045 Paviour-Smith, Martin. 2005. Is it Aulua or education dressed up in kastom?: A report on the ongoing negotiation of literacy and identity in a Ni Vanuatu community. Current Issues in Language Planning 6(2). 224-238. Pike, Kenneth L. 1947. Phonemics. Ann Arbor: The University of Michigan Press. Pritchard, Sonia. 2012. A Cross-Language Study of the production and Perception of Palatalized Consonants. (Thesis). Ottawa. Rangelov, Tihomir, Mary Walworth & Julie Barbour. 2023. A multifaceted approach to understanding unexpected sound change: The bilabial trills of Vanuatu’s Malekula Island. Diachronica 40(3). 384–432. https://doi.org/10.1075/dia.21051.ran Rangelov, Tihomir. 2019. The bilabial trills of Ahamb (Vanuatu): Acoustic and articulatory properties. In Sasha Calhoun, Paola Escudero, Marija Tabain & Paul Warren (eds.), Proceedings of the 19th International Congress of Phonetic Sciences, Melbourne, Australia 2019, 1292-1296. Canberra: Australasian Speech Science and Technology Association Inc. https://assta.org/proceedings/ICPhS2019/papers/ICPhS_1341.pdf Rangelov, Tihomir. 2020. A grammar of the Ahamb language (Vanuatu). Hamilton, New Zealand: University of Waikato. (PhD thesis.) https://hdl.handle.net/10289/14038