The first 10 years (actually, I started working towards my PhD in 1982...)

For us to be able to use language to describe the world around us there must exist an interdependence between the two; at some level of internal (cognitive) representation, variation in the language and variation in the perceived world must map onto the same kind of (abstract) representation. A language is ambiguous when more than one mapping is possible from the language onto either the perceived, or mental, world. Some kinds of mapping are preferred to others, and hence the perceptual difficulty in processing each of the following:

i. The fireman told the woman that he'd risked his life for to install a smoke detector. ii. He'll answer the letter he received tomorrow

In the first case, the sequence "that ..." tends to be interpreted as what was told to the woman (a "complement clause"), and this is incompatible with the subsequent sequence "to install ...". In fact, the "that-clause" should have been interpreted as saying something about which woman he told to install a smoke detector (a "relative clause"). In the second case, an attempt is made to interpret the adverb "tomorrow" as providing information about the immediately preceding verb ("received"), and this is incompatible with the tense of that verb. Instead, "tomorrow" should be associated with "answer".

Until the mid 1980s, researchers believed that these "parsing preferences" were due to more general preferences concerned with the ways in which we group words together; some groupings are more complex than others (as defined by the linguistic analyses that can be applied to these sentences), and the data suggested that we adopt the simplest groupings possible. A fierce debate ensued when evidence was provided suggesting that there are circumstances in which these parsing preferences do not operate (and which questioned whether they operate at all). Altmann and Steedman (1988) and Altmann et al. (1992, 1994) developed further some work initiated by Crain and Steedman (1985), and embedded sentences like the first case above in contexts which introduced two women, just one of whom the fireman had risked his life for. We showed that the difficulty encountered on "to install..." was completely eliminated. On the basis of this and other data, we (and subsequently others) concluded that the interpretation of syntactic structure (i.e. of those groupings of words) proceeds by appeal to the discourse context, and that ambiguity is resolved on the basis of the interdependence between structure and the mental representation of that context.

More recently, we (Altmann, van Nice, Garnham, & Henstra, 1998) extended this approach to the second case above, which is not amenable to the kind of explanation that we had developed to explain the first case. We found that we could prevent the difficulty in these other cases by embedding the sentences in contexts:

iii. Sue's wondering when Sam will answer the letter he received. iv. He'll answer the letter he received tomorrow, she thinks.

The reason the difficulty is avoided is that the first sentence leads the reader to 'expect' an answer to the 'indirect question' - it's 'indirect' because we don't explicitly ask "when will Sam answer the letter?", but instead we frame the question indirectly. In answer to such a question, one could get any of the following:

v. tomorrow vi. He'll answer the letter tomorrow vii. He'll answer the letter he received tomorrow viii. He'll answer the letter he received from his friend tomorrow and so on...

In this work, I argued that the reader predicts an adverbial phrase (e.g. 'tomorrow', 'next week', etc.) at each of the locations shown above, and that this prediction - namely that there will be an adverbial providing information about the act of answering - causes the adverb, when it is eventually encountered, to be interpreted correctly. I am currently writing up a version of the account which explains exactly what form these predictions take. The account is easily extended to explain the earlier kinds of ambiguity mentioned at the top of this page, and can also explain some phenomena concerned with how we establish who did what to whom when interpreting a sentence (and specifically, the interplay between information concerning the syntactic dependencies within the sentence and information concerning the context within which the sentence was uttered).

I've written a relatively recent review of some of this work in the journal Trends in Cognitive Science. And for reasons of space, I've not described my work on Artificial Grammar Learning with Zoltan Dienes and Richard Tunney. I'll add a summary of that work just as soon as I'm able to.

References

Altmann, G.T.M.   (1999). Thematic role assignment in context.   Journal of Memory and Language, 41, 124-145.
Altmann, G. T. M., Garnham, A., & Dennis, Y. (1992). Avoiding the garden path: Eye movements in context. Journal of Memory and Language, 31, 685-712.
Altmann, G. T. M., Garnham, A., & Henstra, J. A. (1994). Effects of syntax in human sentence parsing: Evidence against a structure-based proposal mechanism. Journal of Experimental Psychology: Learning, Memory & Cognition, 20(1), 209-216. 
Altmann, G. T. M., and Steedman, M. J. (1988). Interaction with context during human sentence processing. Cognition, 30(3), 191-238.
Altmann, G.T.M. (1998) Ambiguity in Sentence Processing. Trends in Cognitive Sciences, 1998, 2(4), 146-152.
Altmann, G.T.M., van Nice, K., Garnham, A., & Henstra, J.A. (1998). Late closure in context. Journal of Memory and Language. 38(4), 459-484.
Crain, S. and Steedman, M.J. (1985) On not being led up the garden path: the use of context by the psychological parser. In Natural Language Parsing: Psychological, Computational, and Theoretical Perspectives (Dowty, D., Karttunen, L. and Zwicky, A., eds), pp. 320-358. Cambridge University Press.