Post Doc Interview Talk: Sequential vs. hierarchical syntactic models of human sentence processing
- When:
- Wednesday, October 10, 2012, 02:00 pm - 3:00 pm
- Where:
- 6th Floor Conf. Room (#689)
- Speaker:
- Victoria Fossum
- Description:
Abstract:
Human incremental sentence processing is the process by which we read
a sentence, word-by-word, and ultimately comprehend its meaning. A
central question in sentence processing research is to understand the
precise nature of the linguistic representations that we construct
while comprehending a sentence. Experimental evidence demonstrates
that syntactic structure plays a role in these representations. But
open questions remain about the type of syntactic structure that is
most relevant to the human sentence processing mechanism: is this
syntactic structure sequential or hierarchical? Does it include
lexical information (in which case it is "lexicalized"), or is lexical
information processed independently from the syntactic structure (in
which case the syntactic structure is "unlexicalized")?A previous study (Frank and Bod, 2011) compared unlexicalized
sequential and hierarchical models of human sentence processing, and
found that sequential models explain observed human behavior (e.g. eye
movements) during sentence processing better than hierarchical models.
The authors concluded that the human sentence processing mechanism is
insensitive to hierarchical syntactic structure.We investigate this claim, and find a picture that is more complicated
than the one presented by the previous study. First, we show that
lexicalized syntactic models explain observed human behavior during
sentence processing better than unlexicalized syntactic models.
Second, we consider a broader set of sequential and hierarchical
models, and show that the findings of (Frank and Bod, 2011) do not
generalize to this broader set. Finally, we show why, even within the
set of models considered by (Frank and Bod, 2011), their findings are
not entirely conclusive. Our results indicate that the claim that the
human sentence processing mechanism is insensitive to hierarchical
syntactic structure is premature.Bio:
Victoria Fossum recently completed a one-year postdoc in computational
psycholinguistics at the University of California, San Diego. Prior
to that, she held a one-year postdoc at USC/ISI, where she worked on
morphological and semantic statistical machine translation. She
obtained her Ph.D. in computer science from the University of
Michigan, Ann Arbor, in 2010, but conducted the bulk of her thesis
research in syntax-based statistical machine translation at USC/ISI.
Her research interests include problems in both natural language
processing (in particular, statistical machine translation and
syntactic models of language), and human language processing.Home Page:
