In Rovinj, Croatia, August 2015.
Research

One way to divide up my research is by the kind of methods I use: e.g. several of my projects use responses or reading measures collected in experiments, several others come from a long-term commitment to semantic fieldwork on the Oto-Manguean language Santiago Laxopa Zapotec (Dille’ xhunh Laxup), and some newer projects involve computational cognitive modeling in ACT-R.

Spanning these methods, though, my current research is largely aimed at the three goals below.

Explaining variation in sentence comprehension procedures through skill learning

The best-known patterns of behavior in sentence processing are strong and wide-spread enough to emerge reliably in experimental data. But when we look for variability, we can find it: across tasks, across languages, and across individuals. I'm interested in explaining this variability, and to do so, I have found it useful to consider how an individual's preferred sentence processing procedures may emerge as learned skills, attempts to achieve their goals with minimal difficulty, given their experience.

In my dissertation, I considered how this might help us explain variation in the timing of decisions about meaning across different constructions (e.g. homonymy vs. polysemy) and in different tasks (self-paced reading vs. the Maze task). Unlike static categorizations about what is resolved incrementally and what is delayed, a view of decision-timing as a flexible skill based on the utility and risk of early decisions allows us to explain a wider array of observed behaviors. I have also used this approach to explain cross-linguistic variation, in particular a striking finding that comprehenders of Santiago Laxopa Zapotec avoid early robust predictions for gap locations in relative clauses. On a a skill learning approach, this may be because the language's structure allows for well-performing parsing procedures without gap predictions.

If comprehension behaviors emerge from learning, we might also expect that some variation between individual comprehenders in experiments comes individual differences in learning behavior. This was the focus of my projects at Saarland, where we investigated potential sources for individual differences in pragmatic inferencing in shape-based "reference games". Computational simulations of realistic human performance in ACT-R have given us a proof-of-concept that differences in strategy could emerge from variation in learning parameters, a hypothesis we have validated in subsequent experimental work. I see ACT-R models of strategy selection as a valuable tool for further theoretical refinement across this area of my work.

For instance, see:

Accounting for rich linguistic meaning in sentence processing

We know that humans comprehend sentences incrementally, but we often are missing details in our theories when it comes to the rich details of linguistic meaning. I'm generally very interested in helping fill in these details.

In many cases, these interests regard the timecourse of firm interpretation for pragmatic enrichments like scalar implicature or discourse coherence. Across several experiments in the latter half of my dissertation, I observed readers demonstrating no costly reanalysis for these aspects of meaning, even in cases where their incremental expectations reflect an awareness of the potential enrichments. These patterns may be consistent with widespread and long-term maintenance of several ranked analyses at the level of discourse interpretation. Further follow-ups here in collaboration with Daniel Altshuler and Kelsey Sasaki suggest that firm commitments and costly reanalysis can nevertheless be observed when contexts are exceptionally biased, highlighting the need for representations and processing theories compatible with variably specified interpretations.

In other work, my collaborators and I have followed up on the proposal that the representational boundaries supposed in formal pragmatics may drive the packaging of linguistic content in memory, fueling a set of puzzling processing behaviors for peripheral content like appositive relative clauses. Parts of this research have tried to clarify exactly which pragmatic boundaries would have to be responsible, by expanding the set of natural language constructions we look at for these phenomena. Other parts of this research, in close collaboration with Lalitha Balachandran, have tried to clarify which memory mechanisms could explain the observed effects, when we do see them.

For instance, see:

Relating cognition and communication with theories of linguistic meaning

Through fieldwork on Santiago Laxopa Zapotec, I have become interested in associated motion, a family of verbal constructions that combine lexical verb meaning with an extra event of purposive motion. The fine-grained semantics of these constructions suggest that participant intention plays an important role in the unitization of events in natural language, much as it does elsewhere in event cognition.

How best to divide explanatory burdens between general cognition and theories of linguistic representation is a recurring theme in other projects at the semantics-pragmatics interface, including explaining constraints against redundancy inside the Santiago Laxopa Zapotec noun phrase (where the burden seems to be borne by the grammar), and explaining pressures for stance alignment in natural subjective conversation (where the burden seems to be borne by the social goals of communication).

For instance, see: