Seeing clearly: Frame Semantic, Psycholinguistic, and Cross ...
Seeing clearly: Frame Semantic, Psycholinguistic, and Cross ...
Seeing clearly: Frame Semantic, Psycholinguistic, and Cross ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
CHAPTER 6. FUTURE RESEARCH DIRECTIONS AND CONCLUSIONS 230<br />
several sets of a priori categories, some of them \natural" <strong>and</strong> some of them including<br />
what we really believe to be disparate examples. Presumably a set of categories which are<br />
closer to those which subjects had initially will be easier to apply, <strong>and</strong> will produce higher<br />
agreement on categorization of examples, both among subjects <strong>and</strong> with the experimenter's<br />
categorizations. This would be a procedure similar to that used by Rosch (1973), in which<br />
subjects were taught color names with two di erent sets of exemplars, one containing focal<br />
colors <strong>and</strong> one containing non-focal colors.<br />
We would expect other verbs (<strong>and</strong> nouns) of perception to have similar semantic<br />
structures, but most of them to be somewhat less elaborated, since see is one of the most<br />
important. Thus, we expect to nd at least a physical perception sense <strong>and</strong> a non-physical<br />
cognition sense for verbs such asglimpse, look at, perceive, <strong>and</strong> hear <strong>and</strong> phrases such as<br />
get a glimpse of , get/have a taste of , the smell of , etc. Many of these examples are easy<br />
to construct or to nd in corpora. It might be instructive to run similar experiments using<br />
avariety of these perception words, to test what sense divisions speakers have for them,<br />
whether this two-way distinction is reliable, <strong>and</strong> whether any other senses are similar to<br />
the kinds of elaborations found with see, such as discourse deixis (e.g. Ihear you, brother<br />
`I underst<strong>and</strong> <strong>and</strong> sympathize with both your intended message <strong>and</strong> its implications'),<br />
classi cation (He perceives the proposal as just a delaying tactic), etc.<br />
We would also predict other sorts of prototype e ects in addition to those tested<br />
so far, with more central senses more likely to be spontaneously produced <strong>and</strong> sentences<br />
containing them to be better remembered. St<strong>and</strong>ard experimental techniques could be used<br />
for eliciting \best examples" <strong>and</strong> testing recall.<br />
Corpus Studies <strong>and</strong> Word Sense Disambiguation<br />
From Senses to Statistical Pro les<br />
The frame semantic description of the senses given in Section 2.5 is intended to be<br />
precise <strong>and</strong> to contain all the appropriate information about the syntax <strong>and</strong> semantics of the<br />
arguments of each sense at the appropriate level of generality. But all of this information<br />
is discrete <strong>and</strong> categorical: a sense can occur with exactly the two arguments listed, the<br />
phrase typeoftheseen can be either S n or NP <strong>and</strong> nothing else, the semantic type of the<br />
seer is human (not merely sentient), etc.<br />
It seems clear from a variety ofpsycholinguistic experiments that speakers' knowl-