Overspecification and the Cost of Pragmatic Reasoning about Referring Expressions

摘要: Overspecification and the Cost of Pragmatic Reasoning about Referring Expressions Peter Baumann (baumann@u.northwestern.edu) Department Linguistics Northwestern University, Evanston, IL, USA Brady Clark (bzack@northwestern.edu) Stefan Kaufmann (stefan.kaufmann@uconn.edu) University Connecticut, Storrs, CT, Abstract In current approaches to pragmatic reasoning comprehen- sion production referring expressions is modeled as a result interlocutors’ mutual perspective-taking under additional assumption that speakers try minimize their ar- ticulatory effort or cost. The latter usually not tested instead built into experimental tasks referential language games by artificially restricting set possible available identify refer- ent. We present two game experiments: produc- tion experiment, in which were allowed freely choose expression, comprehension experi- ment replicate earlier findings with our stimuli. Our results show while listeners easily perform reasoning, resort overspecification when prag- matic becomes too high. Keywords: Pragmatics; Language games; Expres- sions; Production; Comprehension Introduction complex situation, like room full people, one may be able single person upon hearing rather short de- scription, “the man hat”, even cases question only hat despite fact he certainly has many other characteristics besides wearing hat. This amazing efficiency human communi- cation made reference use central topic linguistic pragmatics. most influen- tial descriptive account for commu- nication was formulated Grice (1975) form his Maxims Conversation, since then, number propos- als have been provide more quantitative models based on Gricean maxims general principles cognition interaction. prominent examples are game-theoretic strategic (e.g. Benz & Van Rooij, 2007; Franke, 2011; J¨ager, 2011) Bayesian grounded so- cial Frank Goodman, 2012; Goodman Stuhlm¨uller, 2013). both classes it assumed speaker hearer reason each other’s perspec- tives: interpret speaker’s expression referent ‘opti- mal’ perspective speaker, who turn chooses ‘optimal’ hearer’s per- spective, etc. A trivial solution this recursive process explicitly mentions all features intended thus absolutely unambiguous given context. Since such an can hardly qualify efficient, however, above make crucial incurs cost producing utterance, things being equal, preference economic (i.e. shortest least effortful) expression. While theoretically appealing, costs noto- riously hard quantify, articulatory speech 1 negligible (Moon Lindblom, 2003; Locke, 2008). And anecdotally tempted reject notion altogether: people talk lot. However, we do argue against role pro- duction reasoning. Instead, speakers’ behavior cannot explained terms fac- tor alone: Under certain conditions, costlier forms than would required ent, suggests (often termed implicature) itself effortful (like any process) speaker. comprehension, well established effortful: reading, sentences involv- ing implicatures take longer with- out (Hamblin Gibbs, 2003), games, target identification less accurate message involves implicature does Degen 2012). production, hand, option reducing need inference overspecification, i.e. saying strictly necessary. indeed, there plenty empirical evidence speak- ers overspecification. particular, research Koolen, Typological analyses Piantadosi, Tily, Gibson, 2012) typically arguments involving processing restrictions production.