Using distributional data to explore derivational under- markedness: a study of the event/property polysemy in nominalization Fabio Montermini CLLE-ERSS CNRS & Université de Toulouse 2 Jean Jaurès fabio.montermini@univ-tlse2.fr ever, especially in the study of the semantics of Abstract derivational processes, given the pervasiveness and systematicity of such phenomena as polyse- This paper proposes a corpus-based anal- my, semantic underspecification, etc. The first ysis of deverbal suffixed nouns in Italian goal of this talk is thus to present arguments in displaying an ambiguity between a clear favor of an usage-based model of derivational event reading (partenza ‘departure’) and morphology, i.e. an approach in which the prop- a clear property reading (intelligenza ‘in- erties of complex lexemes, and the rules by telligence’). It focuses, in particular on which they are formed, are investigated via a words derived with the suffixes -nza and thorough observation of their real contexts of -zione. Three sets of syntactic contexts use. The perspective adopted here is an exem- for words containing the two suffixes plar-based one, in the sense that morphological (high- and low-frequency -nza words and competence is considered to emerge on the basis high-frequency -zione words) were ex- of the linguistic material speakers are exposed to, tracted from a large corpus of contempo- and that this dynamics can be simulated by tak- rary Italian and coded according to their ing into account large amounts of real usage da- semantic reading. The comparison of the ta. The analysis presented can also be qualified three datasets, on the one side, confirms as distributional, since it is inspired, in its fun- an evolution, already observed in the lit- damental assumption, by distributionalist ap- erature, of -nza from a typically deverbal proaches which are current in semantics (cf. action suffix to a typically deadjectival Lenci, 2008 for an overview), according to property suffix, and, on the other side, which there is a correlation between a unit’s shows that the same ambiguity is ob- meaning and its syntactic distribution. served with -zione nouns, although, un- The second goal of the talk is to provide evi- like the case of -nza, in this case it re- dence in favor of a non-compositional view of mains a marginal feature. The results ob- morphological derivation, according to which the tained show the interest of large-scale semantic properties of complex lexemes cannot empirical observations for the analysis of be simply computed on the basis of the sub- morphological phenomena, and militate elements they contain, but rather on the basis of in favour of a model in which (regular) the lexical relations they enter into. The lack of polysemy should be considered as a con- full isomorphism between the form and the stituting property of derived words. meaning of complex lexemes has been observed 1 Introduction and investigated in many cases and in many lan- guages. These include cases of over-marking, Although having a strong empirical basis is an where an element (e.g. an affix) is present with- important feature of most current studies of mor- out carrying any evident meaning (cf. Roché, phological derivational phenomena, these are 2009, among others, for several examples in often realized on (sometimes very large) series of French), and parallel cases of under-marking, complex words taken in isolation, or on the basis where a relevant semantic differentiation lacks of some examples which are intended to exem- an overt formal counterpart. The existence of the plify the totality of the uses a derived lexeme can latter has been observed since a long time, and is enter into, or at least the most common, ‘un- linked with several other phenomena which are marked’, ones. This approach is reductive, how- well known in the literature on morphology and Copyright © by Fabio Montermini. Copying permitted for private and academic purposes. In Vito Pirrelli, Claudia Marzi, Marcello Ferro (eds.): Word Structure and Word Usage. Proceedings of the NetWordS Final Conference, Pisa, March 30-April 1, 2015, published at http://ceur-ws.org 76 semantics, including (regular) polysemy (Booij, 2010: 78-79), multifunctionality (Luschützky (3) and Rainer, 2013), morphological recycling una partenza / *latitanza / *intelligenza istantanea (Hathout, 2011), etc. Morphological under- ‘an instant departure / lam / intelligence’ marking, precisely, constitutes the main focus of the analysis proposed here. una *partenza / latitanza / *intelligenza di due mesi ‘a two-month departure / lam / intelligence’ 2 The data una *partenza / *latitanza / intelligenza ammirevole In several languages, deverbal nouns present ‘an admirable departure / lam / intelligence’ several instances of systematic polysemy, some of which are well described in the literature (e.g. Roughly, we can distinguish the three types action / result, cf. Rainer, 1996, Bisetto and Mel- above according to four dimensions, as exempli- loni, 2007). In particular, this paper is focused on fied in Table 1. cases of nominalization which, in spite of their frequency, have received less attention (but cf. action punctual bound quantifiable partenza + + + + Kerleroux, 2008 on French) namely deverbal latitanza + – + – nouns displaying an ambiguity between an event intelligenza – – – – and a property reading, as in the following ex- amples for the lexeme vigilanza in Italian: Table 1: Types of deverbal nouns. (1) More specifically, the analysis presented has la polizia ha effettuato una vigilanza continua been carried on on nominalizations containing ‘police guaranteed a continuous control’ the two suffixes -nza and -zione1, which share the property that, when they are constructed on a vs. verb, they are linked, formally and semantically, to its participle (respectively, the present and the la sua vigilanza è calata del 50% past participles) or to the homophonous adjective ‘his/her attention decreased of 50%’ (accogliere / accogliente ⇒ accoglienza ‘ac- ceptance’; educare / educato ⇒ educazione ‘ed- Although Italian is the main focus of this paper, ucation’). In addition, they can also be construct- it should be observed that the same ambiguity ed on an adjective lacking a verbal counterpart can be observed in other Romance languages (cf. frequente ⇒ frequenza ‘frequency’; perfetto (and in English), involving several cognate affix- ⇒ perfezione ‘perfection’), and in this case, base es, such as those derived from Latin -antia, -atio, adjectives more often correspond to an individu- -mentum, -tura. In fact, this ambiguity should al-level predicate. In spite of their similarities, probably be ascribed to a specific property however, derived nouns in -nza and in -zione deverbal suffixes possessed in Latin (cf. (2)), present several important differences. The most since it is not observed with other morphological relevant one is probably the fact that while -nza processes which cannot be directly linked to cor- is mainly attached to stative verbs (cf. Gaeta responding Latin constructions, such as verb- 2002), i.e. verbs which are semantically closer to noun conversions or the (Germanic) deverbal (individual-level) adjectives (cf. Chierchia 1995: suffix -al in English: 177), no such tendency is observed with -zione, which, on the contrary, seems to display a pref- (2) erence for active event verbs. Consequently, Lat.: adaequatio (‘adequacy’), observantia (‘ob- apart from some exceptions (cf. partenza ‘depar- servation’) ture’), the property reading can be virtually ap- plied to all -nza nouns, at least in some of their The polysemy in question can also be linked to uses, while for -zione the situation is reversed: the larger spectrum of meanings that have been 1 observed for deverbal nouns; the typical event In fact, both suffixes may present several different forms reading and the typical property reading, in fact, in surface, whose selection depends on the form of the base they attach to. The forms given are intended to be labels for can be considered as the two poles of a continu- more abstract formal representations (on the formal prob- um which includes the nominalization of more or lems posed by –nza and –zione cf., respectively, Gaeta, less permanent states (cf. Fradin, 2011, 2014): 2002: 127-129; Gaeta 2004, 346-348; Thornton, 1990; Montermini, 2010). 77 most of them do not allow this reading, while struction of property nouns. On the other hand, others accept it, a behavior for which no clear no comparable shift can be observed for -zione. systematicity can be identified: 3 The analysis (4) In order to test the distribution of meanings a. determinazione for -nza and -zione nouns, in particular along the ‘determination’ / ‘determinedness’ event / property divide, I extracted the 61 most frequent lexemes in -anza and -enza (the two educazione possible formal variants)2 in the CorIs3, a large ‘education’ / ‘educatedness’ corpus of written Italian. For each of the lexemes in question, 100 contexts of occurrence were b. istruzione randomly selected, each of which was semanti- ‘instruction’ / *‘educatedness’ cally coded according to its compatibility with one of the two meanings in question. In particu- risoluzione lar, the coding was based on such properties as ‘resolution’ / *‘determinedness’ the possibility of being determined by quantifica- tion or a measure adjective, or the presence / ab- In addition to the general features described sence of temporal boundaries. above, some empirical observations motivate a Figure 1 shows the distribution of meanings deeper large-scale observation of the two deriva- according to the class of the base (verb vs. adjec- tional processes in question. First, for some of tive), and, as expected, a strong correlation be- the -nza nouns displaying an event reading there tween verbal bases and event reading, on the one exists a corresponding noun containing extra side, and adjectival bases and property reading, morphological material denoting a property (cf. on the other, are observed. The diagram also assistenza ‘assistance’ ⇒ assistenzialità); simi- shows that, for the most frequent -nza nouns, the larly, to a past participle can correspond a de- two schemes are more or less equally available. rived noun denoting a property, either in concur- rence with a -zione noun or not (cf. risoluto ⇒ risolutezza, determinato ⇒ determinatezza (vs. determinazione). Second, the observation of real language use shows that lexemes with a typical event meaning can be used as property nouns, and vice-versa, like in the following examples taken from the Web: (5) La produzione basata sulla concorrenza del prezzo tende a tagliare i costi sostenuti dalla produzione basata sulla qualità. Figure 1: Distribution of meanings for the most fre- ‘Production based on low prices (lit. price con- quent -nza nouns according to the class of the base. currence) tends to cut the costs incurred by quali- ty based production’. In order to measure the functioning of this mor- phological process in the speakers’ synchronic Paolo […] era un uomo di estrazione nobile, di competence, the same procedure was applied to grande educazione e istruzione ed estremamente low-frequency words containing -nza in the same religioso e timorato di Dio. corpus (62 lexemes overall having a frequency ≤ ‘St. Paul […] was a man of noble lineage, highly 3, 88 contexts overall). In this case, as shown in educated and very cultivated (lit. of great educa- Figure 2, a strong preference of -nza for adjec- tion and instruction), and extremely religious and tival bases and for the property reading can be God-fearing’. observed. Finally, as shown by Benincà and Penello 2 (2005), and as confirmed by the data I have ana- The lexemes in question range from presenza (frequency lyzed, while nouns with a pure event reading 19,671) to indifferenza (frequency 1,671). Of course, the least was cleared from all lexemes ending in -nza that could were the predominant output for -nza in ancient not be clearly linked to a synchronically existing word. Italian, it is more employed today for the con- 3 http://corpora.dslo.unibo.it/coris_ita.html. 78 References Paola Benincà, and Nicoletta Penello. 2005. Il suffis- so -anza/-enza tra sincronia e diacronia. In M. Grossmann and A.M. Thornton (eds.), La forma- zione delle parole. Atti del XXXVII Congresso in- ternazionale di studi della Società di linguistica italiana. L’Aquila, 25-27 settembre 2003, 69-86. Roma: Bulzoni. Antonietta Bisetto, and Chiara Melloni. 2007. Result nominals: A lexical-semantic investigation. In G. Figure 2: Distribution of meanings for the low- Booij, B. Fradin, E. Guevara, S. Scalise and A. frequency -nza nouns according to the class of the Ralli (eds.), On-line Proceedings of the Fifth Me- base. diterranean Morphology Meeting (MMM5). Fré- jus, 15-18 September 2005. Bologna: Università di This result confirms the observation, mentioned Bologna. above, that in the history of Italian -nza evolved Geert Booij. 2010. Construction Morphology. Oxford from a (mainly) deverbal suffix forming event University Press, Oxford. nouns to a (mainly) deadjectival suffix forming Gennario Chierchia. 1995. Individual-level predicates property nouns. as inherent generics. In: G.N. Carlson and F.J. Pel- Moreover, in order to determine whether the letier (eds.) The Generic Book, 176-223. Chicago / features identified for -nza are specific to this London: University of Chicago Press. construction or belong to deverbal nominal suf- fixes in general, the same procedure was further Bernard Fradin. 2011. Remarks on state denoting no- minalizations. Recherches linguistiques de Vin- applied to a comparable set of words in -zione cennes, 40: 73-99. (including some possible formal variants, like -sione), namely the 61 most frequent forms Bernard Fradin. 2014. La variante et le double. In: F. in the CorIs4. Villoing, S. David and S. Leroy (eds.) Foisonne- ments morphologiques. Études en hommage à Françoise Kerleroux, 109-147. Nanterre: Presses Universitaires de Paris Ouest. Livio Gaeta. 2002. Quando i verbi compaiono come nomi. Un saggio di morfologia naturale. Fran- coAngeli, Milano. Nabil Hathout. 2011. Une approche typologique de la construction des mots. In: M. Roché, G. Boyé, N. Hathout, S. Lignon, M. Plénat, Des unités morphologiques au lexique, 251-318. Paris: Her- mès – Lavoisier. Françoise Kerleroux. 2008. Des noms indisctincts. In: B. Fradin (ed.), La raison morphologique. Hom- Figure 3: Distribution of meanings for -zione nouns. mage à la mémoire de Danielle Corbin, 113-132. Amsterdam / Philadelphia: John Benjamins. As Figure 3 shows, the event / property polyse- my remains marginal for -zione, thus suggesting Alessandro Lenci. 2008. Distributional semantics in that, while this polysemy can be considered as a linguistic and cognitive research. A foreword. Ri- constitutive property of the -nza word formation vista di Linguistica, 20.1: 1-30. pattern, while it constitutes a rare and marked Hans C. Luschützky, and Franz Rainer. 2013. Instru- subpattern for -zione. ment and place nouns: A typological and diachro- nic perspective. Linguistics, 51.6: 1301-1359. Fabio Montermini. 2010. The lexical representation of nouns and adjectives in Romance languages. Re- cherches linguistiques de Vincennes, 39: 135-161. 4 Ranging from amministrazione (frequency 17,139) to previsione (frequency 3,987). 79 Franz Rainer. 1996. La polysémie des noms abstraits: historique et état de la question. In: N. Flaux, M. Glatigny, and M. Samain (eds.), Les noms abstraits. Histoire et théorie. Actes du colloque de Dunkerque (15-18 septembre 1992), 116-126. Vil- leneuve d’Ascq: Presses Universitaires du Septen- trion. Michel Roché. 2009. Pour une morphologie lexicale. Mémoires de la Société de Linguistique de Paris. Nouvelle série, 17: 65-87. Anna M. Thornton 1990/1991. Sui deverbali italiani in -mento e -zione. Archivio glottologico italiano, LXXV (2): 169-207 / LXXVI (1): 79-102. 80