next up previous
Next: Assertion in Explanations Up: Why Recognize Coherence? Previous: Coreference Resolution

The Interpretation of Vague Predicates and the Recognition of Coherence

The difficulty with this argument, however, is that it does not distinguish between example (3) and the following example.

(10)

Ann is so happy! She is tall. She just got a promotion!

Her being tall (except in bizarre contexts) has nothing to do with her happiness or with her promotion. Example (3) is a reasonable thing to say, and (10) is not. Yet the coreference account for ``promotion'' goes through just the same. Example (10) has cohesion but not coherence, whereas example (3) has both.

For a formal account of the intuitive difference between examples (3) and (10), let us first look at another local pragmatics problem--the interpretation of vague predicates, specifically, the vague predicate conveyed by the adjacency of two nouns in a compound nominal, such as ``kitchen light''. The explicit content of the phrase ``kitchen light'' is

(11)

$ kitchen(x) \& nn(x,y) \& light(y) $

In addition to the information conveyed by the word ``kitchen'' and the word ``light'', there is the information that there is some vaguely specified relation between the kitchen and the light. In IA this relation was called nn, and a pragmatic strengthening of this relation is the solution to the compound nominal resolution problem.

Suppose we know that a kitchen is a room:

(12)

$ kitchen(x) \, \supset \,room(x) $

and that the ``in'' relation is a possible nn relation:

(13)

$ in(y,x) \, \supset \,nn(x,y) $

These axioms together with Axiom (5) yield the interpretation that the kitchen light is the light that is in the kitchen. This interpretation is shown in Figure [*].


  
Figure: Interpretation of ``kitchen light''.
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(205,50)(...
...{\raisebox{0pt}[0pt][0pt]{\xipt\rm$kitchen(x)$ }}}
\end{picture}\par\end{figure}

Just as a vague relation is conveyed by the adjacency of the nouns in a compound nominal, so also is a vague relation conveyed by the adjacency of two segments of discourse. We may call this relation CoRel. Thus, when we are interpreting example (3) and we must prove its explicit content, it is not enough to abductively prove (7). That is not all that is conveyed. We must prove

(14)

happy'(e1,A), CoRel(e1,e2,e), promotion(e2,A)

The CoRel predication encodes the information conveyed by the adjacency of the two sentences. We can prove the CoRel predication with Axiom (1) if we find a causal relation between e1 and e2. But that is provided by Axiom (8), used in the establishment of coreference, broadly construed. The fact that axioms get used for more than one problem contributes to the proof's minimality. The proof of (14), that is, the interpretation of example (3) handling both coreference and coherence, is shown in Figure [*].


  
Figure: Interpretation of Example (3) with Coreference and Coherence.
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(335,100)...
... job(e_{2},A) \, \supset \,promotion(e_{2},A)$ }}}
\end{picture}\par\end{figure}

By contrast, in example (10) there is no explanation of the relation conveyed by the adjacency of the successive segments. This is illustrated in Figure [*], where coreference is established but coherence is not. The fact that there is no pragmatic strengthening of the CoRel relations in this example is the formal correlate of our sense of the incoherence of the text. An analogous compound nominal example is ``kitchen job light'', where the nnrelations between ``kitchen'' and ``job'' and between ``job'' and ``light'' would be unexplained.


  
Figure: The Incoherence of Example (10).
\begin{figure}
\par\setlength{\unitlength}{0.0125in} %
\begin{picture}
(395,140)...
... job(e_{2},A) \, \supset \,promotion(e_{2},A)$ }}}
\end{picture}\par\end{figure}

In summary, considerations of coreference and other aspects of local pragmatics can sometimes provide all the information that would be provided by a coherence analysis, but they leave the coherence (or incoherence) itself unexplained.


next up previous
Next: Assertion in Explanations Up: Why Recognize Coherence? Previous: Coreference Resolution
Jerry Hobbs
2000-08-23