ResearchWe study the cognitive processes that underlie language use, and build detailed models of these mechanisms so they can be implemented in computational models. Our research covers the three domains of linguistic knowledge - meaning, form, and the relation between them.
Simulation SemanticsWhat does it mean to understand a piece of language, and what is the nature of the meanings of words and other linguistic elements? Our experimental and theoretical work on Simulation Semantics argues that in order to produce or understand meaningful language, language users run a mental simulation of (that is, mentally imagine) the content of utterances. Running this simulation involves activating the same neural structures that are responsible for perceiving or performing the events described in the utterance, in effect, recreating the internal states associated with perceiving described scenes or performing described actions.
For instance, in understanding a sentence like Mary tossed me a drink, one creates a mental image or simulation of what the scene looks like (size, shape, orientation, trajectory of the drink) and/or what is required to execute the actions involved (body parts used, the amount of force exerted, etc.).
The idea that language drives mental simulation explains
PhonaesthemesWhat are the primitive units that pair form and meaning? Traditionally, this is the role of morphemes, like cat, run and the plural -s. But there exist other correlations in form and meaning that are not compositional and thus are not classical morphemes. These correlations are called phonaesthemes - frequently recurring sound-meaning pairings that are not contrastive morphemes. An example is the English onset gl-, which, like other phonaesthemes, is relatively infrequent in English, except among words with meanings related to VISION and LIGHT, like glimmer, glisten, glow, glimpse and many others. Another well-documented phonaestheme is the English onset sn-, which occurs in a large number of words relating to MOUTH and NOSE, like snore, snack, snout, snarl, and others.
Our experimental work on phonaesthemes shows that not only are phonaesthemes used during language processing, (as seen through priming experiments) but in addition they display priming effects that are very similar to those produced by classical productive morphemes. We conclude from this that form meaning pairings within words are represented as part of the linguistic system, even if they are not morphological.
Embodied Construction GrammarAlong with colleagues at UC Berkeley and elsewhere, we have been developing a model of language representation and use called Embodied Construction Grammar (ECG). ECG includes a computationally precise formalism, but unlike other formal models, is also cognitively oriented in a number of ways.