26.01.2013 Views

Presuppositions in Spoken Discourse

Presuppositions in Spoken Discourse

Presuppositions in Spoken Discourse

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Chapter 6<br />

Two potential problems with Asher & Lascarides’ approach are discussed by<br />

Wilson & Matsui (1998). First, if there are two potential <strong>in</strong>terpretations there may<br />

be alternative anchors each with a plausible coherence relationship. This demands<br />

some sort of order<strong>in</strong>g on coherence relationships or some other method of<br />

dist<strong>in</strong>guish<strong>in</strong>g between them. The second objection is that for ambiguous<br />

sentences, Asher & Lascarides’ approach cannot decide between the two read<strong>in</strong>gs.<br />

For example, if there are two alternative anchors, but each anchor is related to the<br />

bridg<strong>in</strong>g anaphor by the same coherence relationship, then the identification of this<br />

relationship alone will not aid <strong>in</strong> disambiguat<strong>in</strong>g. I do not see why this would be an<br />

objection because if the example is clearly ambiguous, then we should therefore<br />

prefer a method that yields two read<strong>in</strong>gs.<br />

Wilson & Matsui (1998) present an alternative proposal that uses coherence<br />

of <strong>in</strong>terpretation to determ<strong>in</strong>e the anchor and the l<strong>in</strong>k. They use Relevance Theory<br />

(Sperber & Wilson, 1995) as a method of analysis for psychol<strong>in</strong>guistic data first<br />

presented <strong>in</strong> Matsui (1995), where subjects’ <strong>in</strong>tuitions about potentially ambiguous<br />

bridg<strong>in</strong>g examples were tested. These <strong>in</strong>volved cases of two potential NP anchors<br />

to a def<strong>in</strong>ite NP bridg<strong>in</strong>g anaphor where the type of l<strong>in</strong>k was varied as well as the<br />

position of the anchor, factuality, plausibility, etc. In (9) - (10), both New Zealand<br />

and England are potential anchors. However, <strong>in</strong> (9), subjects showed disagreement<br />

about the anchor, whereas <strong>in</strong> (10) they unanimously chose New Zealand, the<br />

factually most plausible anchor.<br />

(9) Kev<strong>in</strong> moved from New Zealand to England. He hates the sheep. (60%<br />

England; 40% New Zealand)<br />

(10) Kev<strong>in</strong> moved from England to New Zealand. He hates the sheep. (100% New<br />

Zealand)<br />

These examples are a challenge for each of the different approaches exam<strong>in</strong>ed <strong>in</strong><br />

Wilson & Matsui (1998). They argue that none of the other approaches account for<br />

<strong>in</strong>tuitions for the preferred anchor shown by this experiment. Instead they believe<br />

that the notion of relevance, as def<strong>in</strong>ed <strong>in</strong> relevance theory could solve many of<br />

these problems, and for reasons of space the reader is referred to their article for<br />

more explanation. It is easy to sympathize with Wilson & Matsui’s criticism of<br />

current proposals and their <strong>in</strong>ability to analyze all examples categorized as bridg<strong>in</strong>g.<br />

However, the proposal by Asher & Lascarides, which utilizes several knowledge<br />

sources, seems to be the one with the greatest potential for be<strong>in</strong>g able to resolve<br />

bridg<strong>in</strong>g <strong>in</strong>ferences.<br />

6.2.3 Empirical work<br />

F<strong>in</strong>ally, some researchers work<strong>in</strong>g <strong>in</strong> computational l<strong>in</strong>guistics have looked at<br />

corpus data, where they have tried to identify and resolve bridg<strong>in</strong>g anaphors. The<br />

most <strong>in</strong>terest<strong>in</strong>g empirical work is the annotation experiments done by Poesio &<br />

Vieira (1998). This is a method that can be used to test and evaluate whether<br />

proposed semantic categories reflect <strong>in</strong>terpreters’ <strong>in</strong>tuitions, and has become a<br />

common methodology with<strong>in</strong> computational l<strong>in</strong>guistics. The standard procedure is<br />

162

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!