Mechanisms Underlying the Comprehension of Language
Michael Tanenhaus
My research focuses on the mechanisms underlying real-time spoken language and reading comprehension. As we are listening or reading, we develop interpretations without waiting until the ends of words, phrases or sentences. This process requires continuous coordination of different types of linguistic and non-linguistic information. Moreover, the sequential nature of the input means that numerous temporary ambiguities routinely arise.
Much of my research has used ambiguity resolution as a vehicle for examining how information is coordinated during real-time language processing, addressing questions such as: (a) how is information from different linguistic subsystems represented, accessed, and combined; (b) how are linguistic and conceptual representations integrated and how do they interact during processing; and (c) to what extent is language processing accomplished by specialized modules that operate according to principles unique to language (as contrasted with more general principles of information integration common to other perceptual and cognitive domains). A particular focus of this work has been on lexical representation and processing within a constraint-based framework. My research examining these issues makes use of a variety of experimental methods and computational models.
Most recently, I have pioneered the use of light-weight head-mounted eye-tracking as a methodology for monitoring spoken language comprehension. Participants follow spoken instructions to manipulate real or computer displayed objects (e.g., Put the candle that's on the towel into the box). Eye movements to the objects are precisely time-locked to relevant information in the instruction as it unfolds, making it possible to study real-time comprehension of spoken language with natural tasks. The methodology is currently being applied to issues in spoken word recognition, syntactic and semantic processing, and discourse comprehension.
Two large-scale projects are underway. The first project examines spoken word recognition in continuous speech, using English as well as artificial languages, and a combination of experimentation and modeling. Recent work has traced the time course of lexical access, resolving outstanding issues about lexical competition and validating a quantitative linking hypothesis between lexical activation and fixations. Work underway is examining the effects of subcategorical phonetic variation on lexical access, integrating classical issues in speech perception into investigations of lexical access. The second project examines syntactic processing and semantic interpretation in natural contexts, including interactive conversation. This work is aimed at bridging two broad traditions in language performance: (1) the "language-as-product" approach which focuses on the recovery (or creation) of context invariant linguistic structure in an utterance and (2) the "language-as-action" approach which views language processing as inextricably bound to a particular time and place, with speakers and hearers cooperating to create context-specific representations.