Meaning to Form: Measuring Systematicity as Information

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1751–1764 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1751 Meaning to Form: Measuring Systematicity as Information Tiago Pimentel♣ Arya D. McCarthy♥ Dami´an E. Blasi♠ Brian Roark♦ Ryan Cotterell♥† ♣Kunumi, ♥Johns Hopkins University, ♠University of Z¨urich & MPI SHH, ♦Google, †University of Cambridge tiago.pimentel@kunumi.com, arya@jhu.edu, damian.blasi@uzh.ch, roarkbr@gmail.com, rdc42@cam.ac.uk Abstract A longstanding debate in semiotics centers on the relationship between linguistic signs and their corresponding semantics: is there an ar- bitrary relationship between a word form and its meaning, or does some systematic phe- nomenon pervade? For instance, does the character bigram gl have any systematic re- lationship to the meaning of words like glis- ten, gleam and glow? In this work, we of- fer a holistic quantification of the systematic- ity of the sign using mutual information and recurrent neural networks. We employ these in a data-driven and massively multilingual approach to the question, examining 106 lan- guages. We find a statistically significant re- duction in entropy when modeling a word form conditioned on its semantic representa- tion. Encouragingly, we also recover well- attested English examples of systematic af- fixes. We conclude with the meta-point: Our approximate effect size (measured in bits) is quite small—despite some amount of system- aticity between form and meaning, an arbitrary relationship and its resulting benefits dominate human language. 1 Introduction Saussure (1916) expounded on the arbitrariness of the sign. Seen as a critical facet of human lan- guage (Hockett, 1960), the idea posits that a sign in human language (a word, in our inquiry) is struc- tured at two levels: the signified, which captures its meaning, and the signifier, which has no mean- ing but manifests the form of the sign. Saussure himself, however, also documented instances of sound symbolism in language (Saussure, 1912). In this paper, we present computational evidence of relevance to both aspects of Saussure’s work. While dominant among linguists, arbitrariness has been subject to both long theoretical debate (Wilkins, 1668; Eco, 1995; Johnson, 2004; Pullum H ( W | V) I (W;V) Wordform (W) Meaning (V) H (W) Wordform (W) Language Model Language Model Figure 1: We use two independent language models to estimate the mutual information between word forms and meaning—i.e. systematicity, as per our definition. The language models provide upper bounds on H(W) and H(W | V), which can be used to estimate I(W;V). This estimate is as good as the upper bounds are tight— see discussion in §3.4. and Scholz, 2007) and numerous empirical and ex- perimental studies (Hutchins, 1998; Bergen, 2004; Monaghan et al., 2011; Abramova and Fern´andez, 2016; Blasi et al., 2016; Gutierrez et al., 2016; Dautriche et al., 2017). Taken as a whole, these studies suggest non-trivial interactions in the form– meaning interface between the signified and the signifier (Dingemanse et al., 2015). Although the new wave of studies on form– meaning associations range across multiple lan- guages, methods and working hypotheses, they all converge on …

Content truncated to 500 words