next up previous
Next: Consequences for the Modelling Up: Assertion in Explanations Previous: Intentionality in the Abductive

Metonymy and Assertion

To resolve this difficulty, we need to look at a third phenomenon from local pragmatics--metonymy. In

(23)

I like to read Proust.

Proust is a person, whereas only texts can be read. We must coerce Proust into some associated text. This can be handled within the syntactic framework sketched above by means of the following kinds of axioms:

(24)

$ Syn(w,x,\ldots) \& rel(x,y) \, \supset \,Syn(w,y,\ldots) $
(25) $ Syn(w,e,\ldots,x,\ldots) \& rel(x,y) \, \supset \,Syn(w,e,\ldots,y,\ldots) $

If we wish to express something involving some x and there is a coercion relation rel between x and y, then we can convey the same information by replacing x with y.

We do not deal here with the very hard problem of just what relations can function as a coercion relation. In the right contexts it can be almost anything. However, as Pustejovsky (1992) argues, coercion relations very often involve causation--the Agentive and Telic qualia, in Pustejovsky's terms. Here we encode possible coercion relations explicitly in axioms such as (28) and (33).

For example (23) three axioms are required.

(26)

$ Proust(y) \, \supset \,write(y,x) \& novel(x) $
(27) $ novel(x) \, \supset \,text(x) $
(28) $ write(y,x) \, \supset \,rel(x,y) $

That is, Proust wrote novels, a novel is a text, and the writerelation is a possible coercion relation.

The interpretation of example (23) is then shown in Figure [*].


  
Figure: Interpretation of Example (23).
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(379,230)...
...]{\xipt\rm$Syn($ \lq\lq read''$,e,\ldots,y,\ldots)$ }}}
\end{picture}\par\end{figure}

Grammatically subordinated material in sentences, such as adjectives modifying nouns, can sometimes carry the primary information of a sentence, and thus be its assertion. For example, in

(29)

I have a sore throat.

it is not the possession of a throat that is being asserted, but the soreness of the throat the hearer already knows the speaker has. This can be viewed as an example of metonymy as well. The explicit assertion of the sentence, the possession, is coerced into the soreness of what is possessed. The possession is related to the throat and the throat is related to the soreness, both by properties that are explicit in the logical form of the sentence and are thus emminently accessible.

This example requires that both possession and soreness be possible coercion relations:

(30)

$ have'(e_{1},z,x) \, \supset \,rel(x,e_{1}) $
(31) $ sore'(e_{2},x) \, \supset \,rel(e_{2},x) $

Moreover, coercion relations compose:

(32)

$ rel(x,y) \& rel(y,z) \, \supset \,rel(x,z) $

Figure [*] now shows the interpretation of example (29). The having e1 is taken as the eventuality conveyed by the verb phrase, but that is coerced into the soreness e2, using as a coercion relation a composite of the having and the soreness. We have not indicated what constraint forces this coercion, but it could be the newsworthiness constraint encoded in Axiom (16), or it could be the weaker constraint that what is said should simply be informative.


  
Figure: Interpretation of Example (29).
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(309,150)...
...m$Syn($ \lq\lq have a sore throat''$,e_{1},\ldots)$ }}}
\end{picture}\par\end{figure}

We now have the machinery we need for dealing with example (3). It can be seen as an instance of metonymy. Just as in example (29) we had to coerce from the having to the soreness, in example (3) we have to coerce from the happiness to the promotion. We need a relation linking them, but the causality provides just such a relation.

(33)

$ cause(e_{2},e_{1}) \, \supset \,rel(e_{1},e_{2}) $

In addition, we need to generalize the Metonymy Axiom (24) to work for coherent segments of discourse and not just grammatical fragments of sentences:

(34)

$ Segment(w,e_{1}) \& rel(e_{1},e_{2}) \, \supset \,Segment(w,e_{2}) $

If w is a coherent segment conveying e1 and e1 is related to e2 by a coercion relation rel, then w can also function as a coherent segment conveying e2.

The pervasive phenomenon of metonymy provides exactly the solution we need to overcome the difficulty pinpointed in Figure [*]. The complete (for this note) interpretation of example (3) is illustrated in Figure [*]. The generalized Metonymy Axiom (34) resolves the conflict between the Explanation pattern (1) and the desideratum of Axiom (16) that an utterance convey newsworthy information; it is precisely the requirement of newsworthiness that forces metonymic interpretation to override the default given by Axiom (1). By Axiom (33) the causal relation between the promotion and the happiness provides the required coercion. That the same causal relation figures in the solutions to the coreference, coherence, and metonymy problems all at once is a virtue of the interpretation, because it is an instance of the exploitation of redundancy that leads to minimal abductive proofs.


  
Figure: Interpretation of Example (3) as Metonymy.
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(356,315)...
...pt}[0pt][0pt]{\xipt\rm$utter(i,u,w_{1} w_{2})$ }}}
\end{picture}\par\end{figure}

Metonymy at the discourse level is a common phenomenon. For instance, Moore and Pollack (1992) provide an example that purports to show that the semantic and pragmatic structures of discourse can sometimes be at odds with each other:

(35)

Come home by 5:00. Then we can go to the hardware store before it closes. That way we can finish the bookshelves tonight.

Pragmatically the principal message is to come home early tonight, making the first sentence the ``nucleus''; semantically, each sentence could be seen as presenting an enabling condition for the state of affairs in the subsequent one, making the final sentence the ``nucleus''. Note, however, that causal chains are reversed by being embedded in desire predicates. If A enables B, then an agent's wanting B will cause the agent to want A. By coercing the physical eventuality of each of the sentences into the wanting of that eventuality, there is a straightforward backward causal chain, and the pragmatic interpretation of the text is derived by Axiom (1) without difficulty.


next up previous
Next: Consequences for the Modelling Up: Assertion in Explanations Previous: Intentionality in the Abductive
Jerry Hobbs
2000-08-23