The TRAINS Project: Natural Spoken Dialogue and Interactive Planning

Home | People | Projects | Publications | Resources


The TRAINS project and its successor, TRIPS, form one of the longest running research efforts on "practical" spoken dialogue: conversation undertaken with a specific task in mind. The TRAINS project had several goals:
  • Undertake a serious study of human-human spoken dialogue
  • Build a series of robust research prototypes untrained users could interact with
  • Use the prototypes as a platform for in-depth research in natural language understanding, mixed-initiative planning, and reasoning about time, actions and events

One of the early achievements of the TRAINS project was the collection of a corpus of task-oriented spoken dialogues, the TRAINS corpus, now available through the LDC.

Over the course of the TRAINS project, a series of demonstration systems were built. The domains for these systems involved routing and scheduling of freight trains, hence the project name. Early systems, such as TRAINS-91 and TRAINS-93 systems, were limited in scope but were important demonstrations of the "doability" of spoken dialogue systems. (These systems are described in James F. Allen et. al., ``The TRAINS Project: A case study in building a conversational planning agent,'' Journal of Experimental and Theoretical AI, 7(1995), 7-48.)

Map of an early TRAINS domain

The TRAINS-95 and TRAINS-96 systems were more robust. (Evaluations were performed of these systems; one is described in Teresa Sikorski and James Allen, ``A Task-Based Evaluation of the TRAINS-95 Dialogue System,'' Proceedings of the ECAI Workshop on Dialogue Processing in Spoken Language Systems, August, 1996.)

Map of a later TRAINS domain

Several dissertations came out of the TRAINS project:

  • Eric Ringger ("Correcting speech recognition errors", 2000)
  • Mark Core ("Dialog parsing: From speech repairs to speech acts", 1999)
  • Lou Hoebel ("A practical temporal reasoning system", 1997)
  • Peter Heeman ("Speech repairs, intonational boundaries and discourse markers: Modeling speakers' utterances in spoken dialog", 1997)
  • Marc Light ("Morphological cues for lexical semantics", 1996)
  • David Traum ("A computational theory of grounding in natural language conversation", 1995)
  • George Ferguson ("Knowledge representation and reasoning for mixed-initiative planning", 1995)
  • Alice Kyburg ("Belief, assertability, and truth: Pragmatic and semantic accounts of vagueness", 1994)
  • Massimo Poesio ("Discourse interpretation and the scope of operators", 1994)
  • Nat Martin ("Applying statistical inference to planning under uncertainty", 1993)

Details about TRAINS are available in several papers. A good overview is the following:

  • James F. Allen, Bradford W. Miller, Eric K. Ringger, and Teresa Sikorski, `` A Robust System for Natural Spoken Dialogue,'' Proceedings of the 1996 Annual Meeting of the Association for Computational Linguistics (ACL'96), June 1996. pp. 62-70.

A more detailed description is:

  • Ferguson, George M., Allen, James F., Miller, Brad W., Ringger, Eric K., "The Design and Implementation of the TRAINS-96 System: A Prototype Mixed-Initiative Planning Assistant", TN96-5, Computer Science Dept., U. Rochester, October 1996.

For more information, visit our Publications site.

TRAINS was funded in part by:

ARPA and the USAF Rome Laboratory under research contract no. F30602-95-1-0025
The Office of Naval Research under research contract no. N0014-95-1-1088
The National Science Foundation under grant number IRI-9623665

Last change: 5 Dec 2000 by stent