NN Compounds in Italian: Modelling Category Induction and Analogical Extension

Abstract

Word compounding, a phenomenon common to all human languages, constitutes an object of interest for many fields in the domain of cognitive science. Both theoretical linguists and typologists – since Pān. ini’s As. t .ādhyāyi (circa 520 BC) – have striven to provide classificatory and explanatory accounts of the common traits of compounds across languages, as well as of their language-specific peculiarities. For cognitive psychologists and psycholinguists, the focus of inquiry has mainly been on crucial aspects of conceptcombination in compound formation and interpretation. More recently, computational linguists have emphasized the critical role played by compounds – as well as by other multi-word expressions – in many Natural Language Processing tasks such as text segmentation, POS-tagging and parsing. Each of these different approaches has generated a considerable body of dedicated literature, theoretical insights and practical methods. However, each sub-field has remained to a great extent isolated from all the others, at times giving rise to a proliferation of overlapping experimental tasks and theoretical accounts of closely-related phenomena. We would like to suggest that a unified analysis of compounding is possible and desirable. Needless to say, any attempt in this direction has to face methodological issues of considerable complexity, but the perspective of obtaining an integrated approach to compound words that is at the same time theoretically sound, cognitively reliable and quantitatively verifiable will – should the attempt prove correct – be worth the effort.