Experiments on the degree of incrementality in human sentence
processing
The goal of the project is to run reading experiments that investigate
the degree of connectedness in human sentence processing. Human
language comprehension has been argued to be incremental
(i.e. language is interpreted on a word-by-word basis). The strongest
version of the incrementality hypothesis is when full connectedness is
also assumed. This means that people incorporate each word into a
single, totally connected syntactic structure, before any following
words. Interpretation of partial sentences is then achieved through
the interpretation of these single syntactic structures; as long as
different bits of sentences are not connected, the relations between
them can't be determined. Methods for obtaining incremental
interpretations are relevant for NLP applications such as speech
recognition and speech-to-speech translation. The main interest in
this project is to determine the degree of connectivity in the human
parser. For example, recent results from psycholinguistics experiments
(Lombardo & Sturt, 2005, see Example 1) indicate that human language
comprehension of coordinated structures is both incremental and
eager. This experiment works on gender mismatch: reading times are
longer in sentence (b) at the point when the untypical gender (herself
for the pilot) is encountered. This finding means that parts of the
second conjunct (put herself) are already connected with syntactic
structure outside the coordinated phrase before the end of the second
conjunct (after "in a very awkward situation") is reached.
1. (a) The pilot embarrassed John and put himself in a very awkward
situation.
(b) The pilot embarrassed John and put herself in a very awkward
situation.
Your task would be to 1) identify syntactic structures for which
connectedness can be determined experimentally and 2) design an
experiment that provides evidence for or against connectedness in this
particular structure. The 3rd step of your project will be to run the
experiments. To do this, you will learn how to conduct studies which
deliver accurate online reading time measures to investigate human
sentence processing. You will receive full training on how to design,
run, and evaluate eye-tracking and self-paced reading experiments.
References:
- Sturt, P. and Lombardo, V. (2005). Processing coordinate structures:
Incrementality and connectedness. Cognitive Science, 29:291-305.
- Mazzei, Lombardo, Sturt (2007). Dynamic TAG and Lexical
Dependencies. In Research on Language and Computation, Springer 5 (3):
309-332, 2007.
- Vera Demberg (2010), PhD Thesis, see under publications.
back to thesis / hiwi topics