Questioning Arbitrariness in Language: a Data-Driven Study of Conventional Iconicity

Proceedings of NAACL-HLT 2016, pages 343–352, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Questioning Arbitrariness in Language: a Data-Driven Study of Conventional Iconicity Ekaterina Abramova Radboud University Nijmegen e.abramova@ftr.ru.nl Raquel Fern´andez University of Amsterdam raquel.fernandez@uva.nl Abstract This paper presents a data-driven investiga- tion of phonesthemes, phonetic units said to carry meaning associations, thus challenging the traditionally assumed arbitrariness of lan- guage. Phonesthemes have received a subs- tantial amount of attention within the cogni- tive science literature on sound iconicity, but nevertheless remain a controversial and un- derstudied phenomenon. Here we employ NLP techniques to address two main ques- tions: How can the existence of phonesthemes be tested at a large scale with quantitative methods? And how can the meaning arguably carried by a phonestheme be induced auto- matically from word embeddings? We develop novel methods to make progress on these fronts and compare our results to previous work, obtaining substantial improvements. 1 Introduction It has long been held in linguistics that since the same concept can be expressed with words whose forms do not resemble each other (e.g., English dog vs. Italian cane vs. German Hund), there is no in- trinsic link between how words sound and what they mean. This feature—arbitrariness—is often con- sidered a hallmark of human language (Saussure, 1916; Hockett, 1959). At the same time, how- ever, over the last decades, mounting evidence from psycholinguistic studies (Markel and Hamp, 1960; Ohala, 1984; Fordyce, 1989) has shown that speak- ers do in fact associate words that contain a particu- lar form with certain meaning—that is, there is a de- gree of iconicity in language in addition to arbitrari- ness, which has been claimed to benefit language learning (Monaghan and Christiansen, 2006; Mon- aghan et al., 2014). Non-arbitrary form-meaning associations come in two basic varieties: primary iconicity (also called ‘true’) and secondary (or ‘conventional’) iconicity. In the former, the sound is thought to directly re- semble the meaning, as in onomatopoeia.1 In the lat- ter, the relationship is a statistical regularity accord- ing to which words that share similar sounds tend to be also similar in meaning, such as a large propor- tion of English words that end with the sound /-æS/ (e.g., crash, slash, mash, trash, dash) being related to destructive action or collision (Hutchins, 1998). The phonetic units that exhibit conventional mean- ing regularities of this latter kind are called phon- esthemes and are the focus of the present paper. In particular, we investigate two main questions: (1) how can the existence of phonesthemes be tested at a large scale by means of a data-driven method? and (2) how can the meaning arguably conveyed by a phonestheme be derived automatically? Phonesthemes are traditionally distinguished from morphemes in being non-compositional. That is, unthinkable can be thought of as being com- posed of morphemes un- (meaning not), think, and -able (meaning capable of), all contributing to the overall meaning of “incapable of being thought” and susceptible of being combined with other units with predictable semantic …

Content truncated to 500 words