MACHINE-READABLE ENTAILMENTS WITH THE ITALIAN PRENDERE CONSTRUCTION EXPRESSING HITTING AND INSULTING EVENTS

MACHINE-READABLE ENTAILMENTS WITH THE ITALIAN PRENDERE CONSTRUCTION EXPRESSING HITTING AND INSULTING EVENTS Ignazio Mauro Mirto Department of Culture e società, Università di Palermo, Italy ABSTRACT The Italian language features a little debated transitive construction with prendere ‘to take/to catch’ in which a prepositional phrase (PP) with an adverbial value occurs mandatorily (e.g. Lui prese a pugni Leo ‘He punched Leo’). Semantically, this construction often implies the use of physical force or verbal offence. In the hitting or insulting event, the notional subject generally is a [+ Human] Agent, whilst the notional direct object generally is a [+ Animate] Affectee ([1]: 4). It can be contended that prendere, which carries no literal meaning, is zero-valent and that the predicate assigning semantic roles is the PP. A computational tool will be illustrated, which automatically performs the following NLP / NLU tasks: it provides a reliable syntactic and semantic representation of the clause type, and it produces machinereadable entailments with three clause types, i.e. sentences with ordinary verbs, support verbs, and the causative verb fare ‘to make/to have’. KEYWORDS Natural Language Processing, Recognizing Textual Entailment, Semantic Role Extraction, Adverbial PPs with a predicative value, Support verbs 1. INTRODUCTION A phraseological unit (PU) of Italian involves one of the most frequent and polysemous verbs, i.e. prendere ‘literally: take, catch, seize, pick up, etc.’ (its closest synonym, i.e. pigliare, occurs less frequently), and a prepositional phrase (PP), as illustrated in (1) and (2) below (it is to be regretted that most examples of the prendere construction proposed below make reference to violence and this is certainly not the author’s choice; the reason lies in the semantics of the construction, often related to injurious events): (1) Max ha preso a bacchettate il compagno di classe Max has taken at stick-hitting.PL the classmate ‘Max hit his classmate with a stick’ (2) I senatori presero Cesare a pugnalate the senators took Caesar at stabs ‘The senators stabbed Caesar’ To the best of our knowledge, no previous studies exist concerning the prendere construction exemplified above (the chapter in [2] dedicated to nouns ending in –ata contains no examples of this construction). Within its mandatory PP, 112 nouns can alternate with each other (this class is not restricted; 80 nouns (approximately 72%) are suffixed with –ate, as a single morpheme). The construction appears to have a collocational nature (cf. [3]: 596). Notice that before International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 17DOI: 10.5121/ijnlc.2022.11302 combining with prendere, the PP is itself a case of grammatical collocation (see [4]). Thus there are grounds to consider the prendere construction as a case of double collocation. Its main characteristics are itemized below: i. there is ‘significant proximity’ ([5]) between the collocating items (i.e. the PP and either prendere or pigliare); ii. regarding the noun following the preposition: (a) it forms a binomial expression with the preposition a ‘to / at’; (b) it must be bare, plural, and unmodified (with a few exceptions, e.g. a male parole, ‘using bad words’); and (c) it designates a ‘process’ (as bacchettate ‘stick-hitting’ and pugnalate ‘stabs’ exemplify in (1) and (2)); iii. the verb prendere (which in Italian is often employed as a support verb, e.g. Max ha preso una decisione ‘Max made a decision’) contributes no meaning as regards semantic roles; and iv. the PP is not a circumstantial (PPs are often optional in Italian); rather, it appears to function predicatively. The purpose of this paper is to illustrate a computational tool, named NLPytaly (cf. [6], [7] and the repository available on GitHub, https://github.com/ignaziomirto2017/nlpytaly/), which can automatically perform the following tasks: first, the identification of the PU (in the active, passive, and certain impersonal forms, e.g. the so-called si passivante); second, the extraction of the participants’ semantic roles (expressed as Cognate semantic roles, see below); third, the identifying of the entailments which the PU establishes with simple sentences deploying two hypernyms: colpire and insultare; and last, the automatic identification of either one-way entailment or paraphrase relations with three distinct clause types sharing the same content morphemes: the ordinary verb construction, the support verb construction, and analytic causatives with the verb fare ‘make / have / let’ (fare causative construction). 2. WHY IS THE PP MANDATORY? When PPs like those in (1) and (2) above combine with prendere (or pigliare), they cannot be removed without modifying the meaning of the sentence, as the comparison between (2) above and (3) below illustrates: (3) I senatori presero Cesare ‘The senators took Caesar’ (e.g. caught / kidnapped him) However, with any other verb the PP is a genuine circumstantial (excluding fare ‘make’: e.g. Fanno a cuscinate ‘They hit each other with pillows’), as the pair (4)-(5) demonstrates: (4) I senatori uccisero Cesare a pugnalate ‘The senators killed Caesar by stabbing him (stabbed Caesar to death)’ (5) I senatori uccisero Cesare ‘The senators killed Caesar’ The very fact that the PP in (2) cannot be removed demonstrates that a pugnalate fulfills distinct functions in (2) and (4). Therefore, this PP is not a circumstantial and, not constituting a noun phrase, it cannot be considered as an argument of prendere. This raises a problem: what is its function? I believe that such a question should be conceived of as being intertwined with the non-literal use of the verb, in that, linguistically, nobody takes, catches, or seizes anything. In (3), prendere does license two arguments and it assigns to each a semantic role, whilst in (2) it evidently does not, otherwise the very same semantic roles would be identified. There are International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 18 therefore valid reasons for analyzing the PP as the genuine predicative core of the clause type and this is the rationale for it being unremovable. The same reasons also hold with the postverbal noun in support verb constructions (e.g. She shot him a scornful glance vs. She shot him, see [6], [8], [9], [10]). That is, the verb prendere is to be interpreted as the indispensable element agreeing with the subject in person and number and carrying such features as e.g. tense and aspect. In addition, its frequent transitivity in literal uses makes available a syntactic grid for the subject and direct object functions. As far as the assignment of descriptive meaning is concerned, however, prendere appears to be inert. 3. MACHINE-READABLE ENTAILMENTS AND PARAPHRASES This section describes one-way entailments and paraphrases which the prendere construction establishes with three clause types: the ordinary verb construction (OVC), the support verb construction (SVC), and the fare causative construction (FCC). The definitions of entailment and paraphrase on which I rely are those provided by [11]: 107-108 (notice that under this definition quantifiers can block the entailments, see [6]: 4-5). 3.1. The ordinary verb construction Both the examples in (1) and (2) can be paraphrased with sentences constructed from ordinary verbs. Such paraphrases obtain by employing the verbs bacchettare ‘hit with a stick’ and pugnalare ‘stab’, respectively: (6) Max ha bacchettato il compagno di classe ‘Max hit his classmate with a stick’ (7) I senatori pugnalarono Cesare ‘The senators stabbed Caesar’ By comparing (1) and (6), and ignoring the null-valent prendere, it can clearly be seen that the content morphemes in (6) are exactly the same as those found in (1) (‘morphemic invariance’, see [12]). Similarly, the content morphemes in (7) are also those which occur in (2). Moreover, the existence of a paraphrastic relation such as that between (1) and (6) means that these sentences entail each other, as sentences (2) and (7) also do. The rationale for the above paraphrases is rather simple: the noun within the PP in e.g. (1) does have a verbal counterpart with the same content morpheme. That is, the noun-verb pair bacchettate : bacchettare exists in Italian. Overall, there are approximately 20 such noun-verb pairs (which is 17% of all nouns occurring in the clause type), and they include diminutives (e.g. schiaffi / schiaffetti ‘slaps / little slaps’) and augmentatives (e.g. spinte / spintoni ‘pushes / big pushes’). It is of use to refer to this first class in the data structure of the tool as Group A. 14 other nouns (12.5%) do not have a verbal counterpart, but one can be easily located, due to their synonymic relationship with a number of nouns in Group A, e.g. ceffoni ‘big slaps’ (*ceffonare does not exist), to be associated with schiaffi ‘slaps’ and then paired with the cognate verb schiaffeggiare ‘to slap’. This latter list will be referred to as Group B. Again, no verbal counterpart with the same content morpheme is available with 69 other nouns (61.5%, Group C). However, with such nouns an entailment relationship can be straightforwardly constructed. This is by means of sentences in which the verb colpire ‘hit’ occurs (this association also holds true with Groups A and B), as the pair formed by (8) and (9) exemplifies. Notice that most PPs formed with the nouns in Groups A, B and C can be replaced with the phrase ‘a colpi di + noun’, with no changes in meaning (cf. [13]). For example, the sentences Lo ha preso a gomitate / a bottigliate ‘He hit him with his elbow / with a bottle’ International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 19 respectively bear the same meanings as Lo ha preso a colpi di gomito, Lo ha preso a colpi di bottiglia. (8) Loro presero l’uomo a cuscinate they took the man at blows.with.a.pillow ‘They beat the man with a pillow’ (9) Loro hanno colpito l’uomo (con un cuscino) ‘They beat the man (with a pillow)’ Finally, the remaining 10 nouns (9%; only two of them have a verbal counterpart, as in Group A) call to mind cases of verbal offence. With such nouns, the counterpart constructed with an ordinary verb can regularly be rendered with insultare ‘to insult’: (10) Loro presero l’uomo a parolacce / insulti they took the man at bad.words / insults ‘They addressed the man using bad words’ (11) Loro insultarono l’uomo ‘They insulted the man’ These nouns can be labelled as Group D. It is worth noticing that with pairs such as (10) and (11) a one-way entailment is observed rather than a mutual entailment: if (10) is true, then (11) must be true. It is also evident that the opposite does not hold, in that someone can be insulted in a number of ways. 3.2. The support verb construction Mutual entailment relationships can be also established between such sentences as the following: (12) Max prese a morsi la mela Max took at bites the apple ‘Max bit the apple’ (13) Max diede dei morsi alla mela Max gave some bites at the apple ‘Max bit the apple’ Sentence (12) differs from the above examples of the construction under scrutiny only on account of its direct object (i.e. la mela ‘the apple’) being [- Animate]. On the other hand, sentence (13) is an instance of the well-known support verb construction (SVC) (see [8]). As the translation in (13) clearly indicates, the verb diede ‘gave’ possesses a blank valence and it carries no literal meaning (see [14]). Let us now focus on the intersection between the data structure which is currently available in NLPytaly for the SVC (approximately 770 predicate nouns such as morso ‘bite’ in (13)) and the four groups in the data structure which were identified in the preceding Section. All the nouns in Group A can also be employed as the predicate noun of a SVC, whilst only 8 nouns in Group B have a counterpart in the SVC. As for Group C, approximately 40 of the nouns included can also be employed in the SVC, whilst only four of those in Group D (insulti ‘insults’, minacce ‘threats’, offese ‘offences’, and pernacchie ‘raspberries’) are included in the SVC. Of International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 20 importance, the entailment between such sentences as (12) and (13) is reciprocal: the truth of the former guarantees the truth of the latter, and vice versa. 3.3. The fare causative construction Unlike in Subsections 3.1 and 3.2, the entailments between an FCC and the prendere construction cannot be mutual, and for compelling reasons. Let us consider an FCC sentence such as Leo fece schiaffeggiare il ragazzo da Max ‘Leo had Max slap the boy’. The truth of this sentence guarantees the truth of at least two other sentences (with which an entailment relationship is therefore established), as illustrated in (14) and (15) below. The former is an additional instance of the construction with prendere, whilst the latter is another FCC in which the cause of the slapping event is the subject Leo: (14) Max prese a schiaffi il ragazzo Max took at slaps the boy ‘Max slapped the boy’ (15) Leo fece prendere il ragazzo a schiaffi da Max Leo made take the boy at slaps by Max ‘Leo had Max slap the boy’ An entailment relationship also characterizes (14) and (15), albeit in one direction only: (15) entails (14), whilst (14) does not entail (15). As suggested above, the reason is simple: if compared to (14), the FCC in (15) features an extra participant (and therefore an additional semantic role), i.e. the causer Leo (namely the sentence subject, licensed by fare, cf. [14]) and this precludes any reciprocal entailment. 4. GENERAL PICTURE, SEMANTIC ROLE EXTRACTION, AND CONCLUDING REMARKS The network of entailments and paraphrases described in Section 3 can be represented diagrammatically as in Figure 1: PRENDERE CONSTRUCTION: INTERACTIONS OVC FCC SVC Figure 1: Machine-readable entailments obtainable with the tool NLPytaly The arrows illustrate the relationships which the prendere construction establishes with the SVC (exemplified with pair (12)-(13)), the FCC (see pair (14)-(15)), and the OVC (see pairs (1)-(6), (2)-(7), (8)-(9), and (10)-(11)). The meaning which NLPytaly automatically extracts is based on a novel type of semantic role, labelled Cognate Semantic Role (CSR, see [6], [7], [9], [10]). For example, with an OVC sentence such as (7) (I senatori pugnalarono Cesare), the NLPytaly tool extracts the CSRs >the one who stabs< and >the one who is stabbed< (these semantic roles are expressed in Italian). International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 21 These cognate semantic roles are subsequently paired with the subject and direct object, respectively. The outcome will therefore be two-fold: (a) ‘the senators are those who stab’, and (b) ‘Caesar is he who is stabbed’. Sentence (2) (I senatori presero Cesare a pugnalate) is initially identified as a prendere construction: prendere is thus ignored and CSRs obtain from the PP having recourse to the verb which is cognate to pugnalate ‘stabs’, i.e. pugnalare ‘to stab’. Thus, the same pairings obtained from (7) are also obtained from (2) and the entailment between (2) and (7) becomes machine-readable, inasmuch as the set of pairings extracted from the former sentence will be identical to that of the latter. This also holds true for the relationship between the prendere construction and the SVC, whilst a one-way entailment (thus devoid of a paraphrase) obtains with the FCC. It is worth noting that no approach which centers on the verb as the only possible assigner of semantic roles can attain the above-mentioned entailments and paraphrases. The rationale for this is that, in the prendere construction (as well as in the SVC), the verb plays no role in the assignment of cognate semantic roles. REFERENCES [1] Fillmore, J. Charles (1968), “The case for case”, in Bach, E. & Harms, R.T. (eds.) Universals in linguistic theory. New York, Holt, Rinehart & Winston, 1-88. [2] Gaeta, Livio (2002), Quando i verbi compaiono come nomi. Un saggio di Morfologia Naturale, Milano, FrancoAngeli. [3] Krishnamurthy, Ramesh (2006), “Collocations”, in Encyclopedia of Language and Linguistics (2nd Edition), R. Brown (Editor-in-chief), Vol. 2, Amsterdam, Elsevier Science, 596-600. [4] Faloppa, Federico (2010). “Collocazioni”, in Enciclopedia dell’italiano Treccani. Available online: https://www.treccani.it/enciclopedia/collocazioni_(Enciclopedia-dell'Italiano)/. [5] Halliday, Michael A. K. & Ruqaiya Hasan (1976), Cohesion in English. London – New York, Hove Longman. [6] Mirto, Ignazio Mauro (2021), “Automatic extraction of semantic roles in support verb constructions”, International Journal on Natural Language Computing (IJNLC) Vol.10, No. 3, 1-10. [7] Mirto, Ignazio Mauro (2022), “Measuring meaning”, in Arai, K. (ed.), Intelligent Computing, LNNS 283, 1054-1067. [8] Gross, M. (1981), “Les bases empiriques de la notion de prédicat sémantique”, Langages, vol. 63, 7-52. [9] Mirto, Ignazio Mauro (2019), “The role of cognate semantic roles: Machine translation support for support verb constructions”, Paper presented at the 13th International Conference NooJ 2019, Hammamet, Tunisia, June 7-9. Available online: (PDF) The role of cognate semantic roles: Machine translation support for support verb constructions | Ignazio Mauro Mirto - Academia.edu. [10] Mirto, Ignazio Mauro (2020), “Natural Language Inference in Ordinary and Support Verb Constructions”, in Dong et al (eds.), Distributed Computing and Artificial Intelligence,17th International Conference, Springer Nature Switzerland, 124-133. [11] Hurford, James R. & Brendan Heasley (1983). Semantics: A coursebook. Cambridge, Cambridge University Press. [12] Harris, Zellig S. (1981), Papers on Syntax, H. Hiż (eds.), London, D. Reidel Publishing Company. [13] Gross, Gaston (1984), “Étude syntaxique de deux emplois du mot ‘coup’“, in Lingvisticae Investigationes, vol. 8: 37-61. International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 22 [14] La Fauci, Nunzio & Ignazio M. Mirto (2003), Fare. Elementi di sintassi. Pisa, ETS. Author Ignazio Mauro Mirto (Ph.D. Cornell University) is Associate Professor of General Linguistics at the University of Palermo. His current work encompasses the area of morphosyntax, information structure, and determiners as inflectional morphemes for the syntactic closure of noun phrases functioning as arguments. oto International Journal on Natural Language Computing (IJNLC) Vol.11, No.3, June 2022 23