Doctoral Dissertation Research: The processing of NPI licensing and intrusion

Suny At Stony Brook, Stony Brook NY

Investigators

Abstract

Language comprehension relies on brain mechanisms which decode incoming signals into meaningful messages in fractions of a second. This involves (i) encoding of the incoming signal, (ii) access/retrieval of information stored in long term memory linking words to both meanings and syntactic function, (iii) temporary maintenance of complex representations of structured phrases, clauses, and sentences, and (iv) a mapping these syntactically structured objects to meanings. How our brains realize this uniquely human capacity, however, is still poorly understood. The present project uses high temporal resolution experimental techniques--including measurements of eye-movements and electrophysiological recordings (EEG) during reading--in order to study in detail the nature and time-course of the various sub-systems involved in extracting sentence-level meaning in real time. A further important feature of this investigation is the focus on cross-linguistic universals and variation by studying how these mechanisms work when processing a well-studied language like English compared to a less well-studied language like Turkish. The specific focus of this work is on the processing of cross-linguistically common expressions known as negative polarity items (NPIs). These include English words like "any" or "ever". Study of these elements in linguistics is important as they expose the way language relates to logical aspects of semantics. NPIs impose demands on the sentences that contain them: they must co-occur with "licensors" like negation (e.g., compare the deviant "Everyone ate *anything" to the well-formed "Nobody ate anything"). Further, NPIs must bear a structural relationship with their licensors in order to be well-formed; it is not enough for a licensor to be merely present (e.g., compare "The man that nobody knows ate *anything" versus "Nobody that the man knows ate anything"). Of central interest to the present project is the fact that these types of dependencies appear to be subject to so-called "intrusion effects" where, in the dynamics of incremental processing, parsing mechanisms appear to sometimes be tricked by the input, resulting in a situation where structurally ineligible licensor/NPI relationships are temporarily processed as if they were licit. This is a species of "grammaticality illusion", where an ungrammatical structure is temporarily judged as well-formed. Such cases have been of interest to language processing researchers, similar to work on illusions in understanding visual cognition, since they appear to reveal important details of how comprehension systems work by understanding cases where they fail or can be "tricked". Competing accounts have been offered in the psycholinguistics literature to explain these intrusion effects with NPIs: (i) cue based retrieval takes intrusion to be a special case of retrieval and dependency resolution error arising from mechanisms posited to underlie a wide variety of types of linguistic dependency processing, and (ii) erroneous pragmatic licensing, which posits that the mechanisms underlying intrusion are specific to NPIs. The present project examines these issues with a special focus on Turkish. Crucially, in Turkish, unlike English/German, NPIs precede their licensors (they realize a prospective/predictive dependency). This work thus systematically examines intrusive NPI licensing cross-linguistically to probe the nature of universals and variation in language processing.