Productivity, Polysemy, and Predicate Indexicality

This paper argues that at least some cases of productive language use, specifically cases of apparent variation in word sense, require a treatment at the conceptual level rather than a lexical semantic solution. It is argued that the lexical semantics should be left underspecified in these cases, and the observed variation in truth-conditions should be attributed to differences in conceptual representation that result from differences in the utterance context. This involves rather drastic changes in our conception of the semantics-pragmatics interface, which are discussed in this paper. Introduction: The Semantics-Pragmatics Interface Language comprehension is concerned in the first instance with linguistic utterances. But, as much recent work emphasizes (e.g. Trueswell & Tanenhaus 2005), language comprehension also makes essential use also of non-linguistic information as it may be available concurrently with the comprehension process. How exactly the processing of linguistic and nonlinguistic information interact is still largely an open issue. In this paper I am exploring an architecture of the language processing system that is decidedly modular in that it clearly distinguishes different subsystems, but that is also incremental and interactional in the sense that it permits current processes to take advantage of information as soon as it becomes available to any one of the subsystems. Similar views were proposed in the language processing literature already in Marslen-Wilson & Tyler (1980) and Altmann & Steedman (1988), and are receiving additional support from recent eye-tracking studies (Hartmann 2005, Karabanov e.a. to app.). The assumption of modularity and incrementality seems to lead to a better fit between theory and observation. But it makes it harder to maintain classic ideas about the semantics-pragmatics interface: It becomes more difficult, in particular, to understand the relation between compositional processes that construct the sentence-semantic content of an utterance, and subsequent pragmatic processes that would finally yield a truth-evaluable content adequate to the utterance context. There may well be additional reasons, and not necessarily all of them empirical, to re-discuss the role of sentence-semantic contents or propositions (Bach 1999, Bosch 1982, Recanati 2002). I will skip these