TRIPS Component: Interpretation Manager

Home | People | Projects | Publications | Resources

TRIPS Architecture | TRIPS Movies | TRIPS Domains | TRAINS vs. TRIPS

The Interpretation Manager guides and controls the interpretation components of the system. When the user produces an utterance, it receives a set of speech acts from the parser which it must 'interpret'. There are four main duties which the Interpretation Manager has.

  1. Detect and interpret indirect speech acts
  2. Coordinate reference resolution
  3. Compute system discourse obligations (and recognize user grounding)
  4. Conversion from linguistic knowledge representation to domain specific knowledge representation.
Each of these will be discussed individually.

Indirect Speech Acts

Parser sends a set of speech acts from the utterance to the Interpretation Manager. These speech acts, however, are only the surface speech acts from the utterance. Oftentimes a surface speech act has a deeper intended or indirect speech act. Take, for example, the utterance, "Can you pass the salt?", which on the surface looks like a QUERY speech act (i.e. asking about your ability to pass the salt). However, the indirect speech act is most likely a REQUEST (i.e. a request to pass me the salt).

The Interpretation Manager detects and interprets indirect speech acts through a two step process mentioned in (Hinkelman and Allen '89). The first step generates possible indirect speech acts based on linguistic syntactic and semantic patterns. (Utterances like "Can you please X?" may possibly be REQUESTs, although this is done at a syntactic and semantic level). These possibilities are then filtered by intention recognition done in conjunction with the Task Manager (also see Knowledge Representation Conversion below). The Interpretation Manager sends each of the postulated indirect speech acts to the Task Manager and asks whether this interpretation would 'make sense' given this specific domain and planning context. Each of these postulates are then scored by the Task Manager which allows the Interpretation Manager to find the correct interpretation of the surface speech act.

Reference Resolution Coordination

Entities in the speech acts that come in from the Parser are represented semantically as object descriptions. The system must ground these objects to something in the 'real world', which can then be used in formulating plans, executing actions, etc. The Interpretation Manager sends requests to the Reference Component to determine the referents of entities in the utterance.

Discourse Obligations

The Interpretation Manager has two duties with respect to discourse obligations. First, it must decide, based on the utterance, what discourse obligations the system now has as a result of the utterance. These new obligations are sent to the Discourse Context. Secondly, the system must decide which of the outstanding discourse obligations the user has fulfilled based on the utterance, which are then removed from the Discourse Context. This second duty includes determining whether or not the user has grounded the system's last utterance. (for more information about discourse obligations and grounding please see the Discourse Context Component)

Knowledge Representation Conversion

In conjunction with its
indirect speech acts responsibilities, the Interpretation Manager must convert the speech act semantics from a linguistic knowledge representation (produced by the Parser) to a domain specific, problem-solving knowledge representation (used by the Task Manager and the Behavioral Agent). For example, for the utterance "Take truck 1 to Pittsford", the linguistic semantics might be some sort of predicate take1.
        :agent SYSTEM
        :theme TRUCK1
        :goal PITTSFORD)
The domain specific, problem-solving knowledge representation would correspond to some action in the domain (move) with its corresponding known parameters. For example:
        :vehicle TRUCK1
        :to-loc PITTSFORD
        :route ?r)
Here the ?r is a variable, since the route to take was not explicitly mentioned in this utterance.

This problem-solving representation is what is then used in both the Task Manager (for intention recognition) and the Behavioral Agent (for accomplishing what the user asked the system to do).

Message Traffic

Current Researchers

Relevant Papers

  • Hinkelman, E. and J. F. Allen, "Two Constraints on Speech Act Ambiguity", Proceedings of the Association for Computational Linguistics (ACL), Vancouver, Canada, 1989.

Last change 07 Nov 2000 by blaylock