We argue that grammatical analysis is a viable alternative to concept spotting for processing spoken input in a practical spoken dialogue system. We discuss the structure of the grammar, and a model for robust parsing which combines linguistic sources of information and statistical sources of information. We discuss test results suggesting that grammatical processing allows fast and accurate processing of spoken input.

This paper describes ongoing development work on the spoken dialogue system, WAXHOLM, providing information on boat traffic in the Stockholm archipelago. The dialogue control and the natural language parser are implemented in an integrated, knowledge-based probabilistic language model. The recognition process is based on neural nets, A* lexical search, and a candidate reordering module. Speech synthesis for spoken response has been enhanced by the display of a synthetic, animated face. Application-specific data have been collected with the help of Wizard-of-Oz techniques.

In this paper, we compare the search effort of the word conditioned and the time conditioned tree search methods. Both methods are based on a time-synchronous, left-to-right beam search using a treeorganized lexicon. Whereas the word conditioned method is well known and widely used, the time conditioned method is novel in the context of 20 000--word vocabulary recognition. We extend both methods to handle trigram language models in a one--pass strategy. Both methods were tested on a train schedule inquiry task (1 850 words, telephone speech) and on the North American Business (Nov.'94) development corpus (20 000 words).

In the course of a "man-machine" dialogue, the system's belief concerning the user's intention is continuously being built up. Moreover, restricting the discourse to a narrow application domain further constrains the variety of possible user reactions. In this paper, we will show how these knowledge sources may be utilized in a stochastic framework to improve speech understanding. On field-test data collected with our automatic exchange board prototype PADIS, a relative reduction of attribute errors by 27% has been obtained.

This thesis presents work in three main directions of the automatic speech recognition field. The work within two of these -- dynamic decoding and hybrid HMM/ANN speech recognition -- has resulted in a real-time speech recognition system, currently in use in the human/machine dialogue demonstration system WAXHOLM, developed at the department. The third direction is fast unsupervised speaker adaptation, where "fast" refers to adaptation with a small amount of adaptation speech. The work in

In 1995 our department was involved in two projects in the field of continuous speech recognition. The main aim of these two strongly related projects was the development of basic technology that can be used to build advanced telephone-based systems for providing information about public transport. A short description of the work carried out within these projects is provided in the present article. 1. Introduction During the last decade the performance of spoken dialogue systems has improved substantially. At the moment, the quality of these systems seems to be able to support a number of simple practical tasks in small and clearly delimited domains. As a result, much effort is spent nowadays to develop prototype telephone-based information systems in different countries. These systems are reminiscent of the well-known Air Travel Information System (ATIS) task that has been a focal point in the American ARPA-project. In Europe two MLAP (Multi-Lingual Action Plan) projects concerning p...

This paper describes an architecture that supports adaptive mixed-initiative dialogue. It is based on a generalisation of the form-filling paradigm. Rather than a flat slot structure, we use a hierarchy that contains slots at various levels of abstraction. Along with the slot hierarchy, a question hierarchy is defined that allows for adaptive mixed-initiative dialogue. Depending on the success or failure of certain questions, the system can zoom-in to more detailed questions, or zoom-out to higherlevel questions.

Browsing documents in the World Wide Web usually offers little opportunity for real interaction, as user input is often limited to mouse clicks and to filling out forms. By adding language technology, the gap between navigation and interaction can be bridged, which should lead to a more symmetric communicative setting. It means that free typed or spoken utterances can be used as means of access to documents on the Web, in addition to mouse clicks and typed strings as they are used to fill out forms, e.g. in search engines. A crucial factor in system design is a sound pre-design study, which investigates how users will use typed and spoken queries to complement existing functionality. In this paper, we are going to describe a Wizard-of-Oz (WoZ) study which has shown that users tend to prefer multi-modal input over traditional input modes in search tasks concerning German newspaper texts, and that they were generally faster in completing the tasks whenever they were allowed to use free typed and spoken input. Additionally, users were grouped according to previous experience with search engines and the Internet. It can be shown that non-expert users expressed a stronger preference for multi-modal interaction than expert users and that, with multi-modal interaction, task completion times were reduced more significantly among non-experts than among experts users.

ion (Johnson and Dorre, [39]) ffl x(A,B,f(A,B),g(A,h(B,i(C)))) =) x(A,B,f(,),g(,)) ffl parsewithweakening(Cat,P0,P,E0,E) :- weaken(Cat,WeakenedCat), parse(WeakenedCat,P0,P,E0,E), Cat=WeakenedCat. ffl Really helps! Ambiguity Packing ffl A parser should not construct all parse trees (exponential) ffl Instead, a compact representation of all such parse trees are constructed -- grammar [42, 9] -- parse forest [76] -- packed structures [3] ffl Here: for every `result item' keep track of the lexical entry and references of other result items that were used to create it ffl Results in a lexicalized tree substitution grammar ffl which generates the input sentence with all its parse trees Bottom-up Inactive-chart Parser Item form: [i;X; j] Axioms: Goals: [0;S;n] Inference Rules: Scan [q i ;wi; qi+1 ] Complete [q k ;X1; q k 0][q k 0;X2; q k 00] : : : [q m0;Xl; qm] [q k ;X0; qm] X0 !X1:::Xl Bottom-up Inactive-chart Parser Inference Rules: Scan [q i ;wi; qi+...

In 1995 our department was involved in two projects in the field of continuous speech recognition. The main aim of these two strongly related projects was the development of basic technology that can be used to build advanced telephone-based systems for providing information about public transport. A short description of the work carried out within these projects is provided in the present article. 1.