July 22, 2007

Wait, wait, don't tell me

In Prague last month, the Association for Computational Linguistics honored Lauri Karttunen with its Lifetime Achievement Reward for his contributions to the field. In his presentation of the award, Mark Steedman mentioned Lauri's work on discourse semantics, unification-based parsing, and finite-state-based approaches to morphology and syntax, any one of which would have justified the award by itself. But before he elected to sully his hands with computation, Lauri had a notable career as a linguistic semanticist, where he made a specialty of raising puzzles about topics like presupposition and reference and anaphora that researchers are still scratching their heads over several decades later (try doing a Google search on "Karttunen plugs OR paycheck"). At the end of his ACL acceptance speech, Lauri suggested that some of those semantic issues are now becoming relevant to NLP, with the advent of search engines that actually make use of semantic processing in addition to simple string matching, so as to be able to draw textual inferences. By way of example, he mentioned the classification of complement constructions, and then, faithful to his past practice, he left the audience with a little conundrum to puzzle over, which I hereby pass on to you, gentle readers. Riddle me this one, and no fair peeking at his ACL slides.

The construction didn't wait to is ambiguous. Here are a couple of examples from Google to to illustrate the ambiguity.

(21) a. Deena did not wait to talk to anyone. Instead, she ran home.
b. It hurt like hell, but I'm glad she didn't wait to tell me.

(21a) implies Deena did not talk to anyone. But (21b) implies She told me something right away.

Question 1: How does it come about that X didn't wait to do Y means either that X did Y right away or that X didn't do Y at all?

When you look at examples with didn't wait to in their full context, it is nearly always possible to tell which of the two meanings the author has in mind. In (21a), for instance, the negative polarity item anyone and the word instead are telltale indicators. In (21b), the cataphoric pronoun it indicates that a telling event took place. I am sure that it is possible to learn to pick the intended meaning by statistical techniques. But statistics alone will not give you an answer to Question 1, nor will it solve the related problem in Question 2.

Question 2: Why is it not possible to translate expressions such as Neil didn't wait to take off his coat to other languages in a way that preserves the ambiguity the sentence has in English?

In languages such as Dutch, Finnish, French, German, Hungarian, and Japanese among others, it is of course possible to express the two meanings of X did not wait to Y but not in one and the same sentence.

Posted by Geoff Nunberg at July 22, 2007 02:28 PM