Comparing Theories of Speaker Choice Using a Model of Classifier Production in Mandarin Chinese

Proceedings of NAACL-HLT 2018, pages 1997–2005 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Comparing Theories of Speaker Choice Using a Model of Classifier Production in Mandarin Chinese Meilin Zhan Massachusetts Institute of Technology 77 Massachusetts Avenue Cambridge, MA 02139 meilinz@mit.edu Roger Levy Massachusetts Institute of Technology 77 Massachusetts Avenue Cambridge, MA 02139 rplevy@mit.edu Abstract Speakers often have more than one way to ex- press the same meaning. What general princi- ples govern speaker choice in the face of op- tionality when near semantically invariant al- ternation exists? Studies have shown that op- tional reduction in language is sensitive to con- textual predictability, such that the more pre- dictable a linguistic unit is, the more likely it is to get reduced. Yet it is unclear to what ex- tent these cases of speaker choice are driven by audience design versus toward facilitating production. Here we argue that for a different option- ality phenomenon, namely classifier choice in Mandarin Chinese, Uniform Information Density and at least one plausible variant of availability-based production make opposite predictions regarding the relationship between the predictability of the upcoming material and speaker choices. In a corpus analysis of Man- darin Chinese, we show that the distribution of speaker choices supports the availability-based production account, and not Uniform Informa- tion Density. 1 Introduction The expressivity of natural language often gives speakers multiple ways to convey the same mean- ing. Meanwhile, linguistic communication takes place in the face of environmental and cognitive constraints. For instance, language users have lim- ited memory and cognitive resources, the environ- ment is noisy, and so forth. What general princi- ples govern speaker choice in the face of alterna- tions that are (nearly) semantically invariant? To the extent that we are able to provide a general answer to this question it will advance our funda- mental knowledge of human language production. Studies have shown that alternations are very often sensitive to contextual predictability. For well-studied cases of optional REDUCTION in lan- guage, the following trend is widespread: the more predictable a linguistic unit is, the more likely it is to get reduced. Predictable words are phonetically reduced (Jurafsky et al., 2001; Bell et al., 2009; Seyfarth, 2014) and have shorter lexical forms (Pi- antadosi et al., 2011), and optional function words are more likely to be omitted when the phrase they introduce is predictable (Jaeger, 2010). Yet it is unclear to what extent speakers’ choices when faced with an alternation are made due to audi- ence design or to facilitate production. For exam- ple, the above pattern of predictability sensitivity in optional reduction phenomena is predicted by both the Uniform Information Density (UID) hy- pothesis (Levy and Jaeger, 2007), a theory which that the speaker aims to convey information at a relatively constant rate and which can be moti- vated via considerations of optimality from the comprehender’s perspective (e.g., Smith and Levy, 2013), and by the speaker-centric availability- based production hypothesis (Bock, 1987; Ferreira and Dell, 2000), which …

Content truncated to 500 words