Nate Blaylock

A problem solving model for collaborative agents

This paper describes a model of problem solving for use in collaborative agents. It is intended as a practical model for use in implemented systems, rather than a study of the theoretical underpinnings of collaborative action. The model is based on our experience in building a series of interactive systems in different domains, including route planning, emergency management, and medical advising. It is currently being used in an implemented, end-to- end spoken dialogue system in which the system assists a person in managing their medications. While we are primarily focussed on human-machine collaboration, we believe that the model will equally well apply to interactions between sophisticated software agents that need to coordinate their activities.

Generating artificial corpora for plan recognition

Corpora for training plan recognizers are scarce and difficult to gather from humans. However, corpora could be a boon to plan recognition research, providing a platform to train and test individual recognizers, as well as allow different recognizers to be compared. We present a novel method for generating artificial corpora for plan recognition. The method uses a modified AI planner and Monte-Carlo sampling to generate action sequences labeled with their goal and plan. This general method can be ported to allow the automatic generation of corpora for different domains.

Managing Communicative Intentions with Collaborative Problem Solving

Dialogue systems need to be able to understand a user’s communicative intentions, reason with those intentions, form their own communicative intentions, and realize those intentions with actual language to be uttered to the user. Oftentimes in dialogue systems, however, what these communicative intentions actually correspond to is never clearly defined. We propose a descriptive model of dialogue, based on collaborative problem solving, which defines communicative intentions as attempts to modify a shared collaborative problem-solving state between the user and system. Modeling dialogue at the level of collaborative problem solving allows us to model a wider array of dialogue types than previous models, including the range of collaboration paradigms (master-slave to mixedinitiative) and interaction types (planning, execution, and interleaved planning and execution). It also provides a definition for utterance-level communicative intentions for use within a dialogue system.

Synchronization in an asynchronous agent-based architecture for dialogue systems

Most dialogue architectures are either pipelined or, if agent-based, are restricted to a pipelined flow-of-information. The TRIPS dialogue architecture is agent-based and asynchronous, with several layers of information flow. We present this architecture and the synchronization issues we encountered in building a truly distributed, agent-based dialogue architecture.

A corpus of clinical narratives annotated with temporal information

Clinical reports often include descriptions of events in the patients medical history, as well as explicit or implicit temporal information about these events. We are working towards applying deep Natural Language Processing tools towards understanding such narratives. This requires both the extraction and classification of the relevant events, and the placing of those events in time, or at least in relation to one another. Although several corpora of news data exist that have been annotated using the TimeML schema, similar corpora of clinical reports are not readily available. In this paper we report on the design of a small corpus and the annotation schema we developed, based on data from the fourth i2b2/VA challenge. These data include, among others, annotations for medical problems, tests, and treatments in clinical reports from several healthcare institutions. We have selected a subset of clinical reports and added annotations similar to those used in the TempEval tasks for the annotation of events, time expressions and temporal relations for the news domain. The annotations have been made freely available to the research community.

Talking through procedures: an intelligent space station procedure assistant

We present a prototype system aimed at providing spoken dialogue support for complex procedures aboard the International Space Station. The system allows navigation one line at a time or in larger steps. Other user functions include issuing spoken corrections, requesting images and diagrams, recording voice notes and spoken alarms, and controlling audio volume.

TESLA: A Tool for Annotating Geospatial Language Corpora

In this paper, we present The gEoSpatial Language Annotator (TESLA)--a tool which supports human annotation of geospatial language corpora. TESLA interfaces with a GIS database for annotating grounded geospatial entities and uses Google Earth for visualization of both entity search results and evolving object and speaker position from GPS tracks. We also discuss a current annotation effort using TESLA to annotate location descriptions in a geospatial language corpus.

The SAMMIE System: Multimodal In-Car Dialogue

The SAMMIE system is an in-car multi-modal dialogue system for an MP3 application. It is used as a testing environment for our research in natural, intuitive mixed-initiative interaction, with particular emphasis on multimodal output planning and realization aimed to produce output adapted to the context, including the drivers attention state w.r.t. the primary driving task.

Semantic Annotation of Street-Level Geospatial Entities

In this paper, we describe the PURSUIT Corpus -- an annotated corpus of geospatial path descriptions in spoken natural language. PURSUIT includes the spoken path descriptions along with a synchronized GPS track of the path actually taken. Additionally, we have manually annotated geospatial entity mentions in PURSUIT, mapping them onto point entries in several geographic information system databases. PURSUIT has been made freely available for download.

Hierarchical Goal Recognition

This chapter discusses hierarchical goal recognition: simultaneous online recognition of goals and subgoals at various levels within an HTN-like plan tree. We use statistical, graphical models to recognize hierarchical goal schemas in time quadratic with the number of the possible goals. Within our formalism, we treat goals as parameterized actions, necessitating the recognition of parameter values as well. The goal schema recognizer is combined with a tractable version of the Dempster-Shafer theory to predict parameter values for each goal schema. This results in a tractable goal recognizer that can be trained on any plan corpus (a set of hierarchical plan trees). Additionally, we comment on the state of data availability for plan recognition in general and briefly describe a system for generating synthetic data using a mixture of AI planning and Monte Carlo sampling. This was used to generate the Monroe Corpus, one of the first large plan corpora used for training and evaluating plan recognizers. This chapter also discusses the need for general metrics for evaluating plan recognition and proposes a set of common metrics.