Stephen Isard

Automatically predicting dialogue structure using prosodic features

Spoken dialogue systems need to track dialogue structure in order to conduct sensible conversations. In previous work, we used only a shallow analysis of past dialogue in predicting the current dialogue act. Here we show that a hierarchical analysis of dialogue structure can significantly improve dialogue act recognition. Our approach is to integrate dialogue act recognition with speech recognition, seeking a best overall hypothesis for what words have been spoken and what dialogue act they represent, in the light of both the dialogue history so far and the current speech signal. A useful feature of this approach is that intonation can be used to aid dialogue act recognition by combining it with other information sources in a natural way.

Using prosodic information to constrain language models for spoken dialogue

We present work intended to improve speech recognition performance for computer dialogue by taking into account the way that dialogue context and intonational tune interact to limit the possibilities for what an utterance might be. We report on the extra constraint achieved in a bigram language model, expressed in terms of entropy, by using separate submodels for different sorts of dialogue acts, and trying to predict which submodel to apply by analysis of the intonation of the sentence being recognised.

Evaluation of speech synthesis techniques in a comprehension task

Abstract Six types of speech synthesis were evaluated for comprehensibility: standard linear predictive coding analysis/ resynthesis; pitch synchronous analysis/resynthesis; pitch synchronous multi-pulse analysis/resynthesis; and three PSOLA (pitch synchronous overlap-and-add) techniques. The relative comprehensibility of the synthesis types was tested by using the synthesised speech to convey information that subjects needed in order to perform a diagram-based multiple-choice task.

Modal Tic-tac-toe

The work we are doing began as an attempt to find a formal definition of truth for a limited class of English sentences. Guided by Tarski’s definition of truth for formalised languages, and by Chomsky’s conception of natural language syntax, we wanted to take a subset of English suitable for discussing a model universe, and to define in a intuitively satisfying way what it would mean for a sentence to be true of that universe, in terms of the syntactic structure of the sentence and the referents of the individual words. We did not feel committed to any specific grammar of English, but we were — and are — strongly prejudiced to the view that English syntax must serve some purpose in the process of communication, and we wanted our scheme to assign some function to as much as possible of the syntax we would use.

Three open questions in the theory of one-symbol Smullyan systems

are symbols in P. A theorem of S is any string which is either an axiom of S or is derivable from the axioms by a finite number of applications of (a) uniform substitution of non-null strings in K* for variables, and (b) modus ponens. A string ~ ~ K* is said to be B-generated by S if Bet is a theorem of S, and a set L is said to be generated by S if there is a B ~ P such that L is the set of all B-generated strings. These definitions are based on the presentation of elementary formal systems in Smullyan 1961.