Finding Non-Arbitrary Form-Meaning Systematicity Using String-Metric Learning for Kernel Regression

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 2379–2388, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Finding Non-Arbitrary Form-Meaning Systematicity Using String-Metric Learning for Kernel Regression E. Dar´ıo Guti´errez UC San Diego edg@icsi.berkeley.edu Roger Levy MIT rplevy@mit.edu Benjamin K. Bergen UC San Diego bkbergen@ucsd.edu Abstract Arbitrariness of the sign—the notion that the forms of words are unrelated to their meanings—is an underlying assumption of many linguistic theories. Two lines of research have recently challenged this as- sumption, but they produce differing char- acterizations of non-arbitrariness in lan- guage. Behavioral and corpus studies have confirmed the validity of localized form-meaning patterns manifested in lim- ited subsets of the lexicon. Meanwhile, global (lexicon-wide) statistical analyses instead find diffuse form-meaning system- aticity across the lexicon as a whole. We bridge the gap with an approach that can detect both local and global form- meaning systematicity in language. In the kernel regression formulation we in- troduce, form-meaning relationships can be used to predict words’ distributional semantic vectors from their forms. Fur- thermore, we introduce a novel metric learning algorithm that can learn weighted edit distances that minimize kernel regres- sion error. Our results suggest that the English lexicon exhibits far more global form-meaning systematicity than previ- ously discovered, and that much of this systematicity is focused in localized form- meaning patterns. 1 Introduction Arbitrariness of the sign refers to the notion that the phonetic/orthographic forms of words have no relationship to their meanings (de Saussure, 1916). It is a foundational assumption of many theories of language comprehension, production, acquisi- tion, and evolution. For instance, Hockett's (1960) influential enumeration of the design features of human language ascribes a central role to arbi- trariness in enabling the combination and recom- bination of phonemic units to create new words. Gasser (2004) uses simulations to show that for large vocabularies, arbitrary form-meaning map- pings may provide an advantage in acquisition. Meanwhile, modular theories of language compre- hension rely upon the duality of patterning to sup- port the independence of the phonetic and seman- tic aspects of language comprehension (Levelt et al., 1999). Quantifying the extent to which the ar- bitrariness principle actually holds is important for understanding how language works. Language researchers have long noted excep- tions to arbitrariness. Most of these are patterns that occur in some relatively localized subset of the lexicon. These patterns are sub-morphemic because, unlike conventional morphemes, they cannot combine reliably to produce new words. Phonaesthemes (1930) are one example. phonaestheme is a phonetic cluster that recurs inA many words that have related meanings. One no- table phonaestheme is the onset gl-, which occurs at the beginning of at least 38 English words re- lating to vision: glow, glint, glaze, gleam, etc. (Bergen, 2004). At least 46 candidate phonaes- themes have been posited in the linguistics liter- ature, according to a list compiled by Hutchins (1998). Iconicity is another violation of arbitrari- ness that can lead to non-arbitrary local regular- ities. Iconicity occurs when the …

Content truncated to 500 words