You are here

A Text-Chat Based Natural Language Interface Toolkit

Description:

OBJECTIVE: Leverage results of synthetic teammate (intelligent agent) research to develop a text-chat based natural language interface toolkit that will facilitate the creation of constructive entities capable of functioning as teammates in training simulations. DESCRIPTION: Text-chat based communications are becoming ever more common in Air Force operations environmentsespecially Unmanned Aerial Vehicle (UAV) and Air and Space Operations Center (AOC) operations. To support the training of airmen in these environments, training simulations need to incorporate constructive entities functioning as synthetic teammates (intelligent agents) which are capable of text-chat based communications with human teammates who are being trained. To facilitate the development of language capable constructive entities, a Natural Language Interface (NLI) toolkit which leverages prior synthetic teammate research (Ball et al., 2010, Rodgers et al., 2011, in press) is needed. Starting with a language capable synthetic teammate developed for a specific domain (Ball et al., 2010), extract out generalizable elements and use the generalizable elements as the basis for creation of an NLI toolkit that facilitates development of language capable constructive entities for other domains. Key technologies to be leveraged include state of the art language analysis (Ball, 2011), situation modeling (Rodgers, et al., 2011; in press) and language generation capabilities. It is not expected that existing synthetic teammate technologies will be entirely sufficient for research and development of the NLI toolkit. An important part of the research will include the identification of technological gaps, and research and development of solutions to close the identified technological gaps. Adoption of state-of-the-art computational linguistic techniques to fill identified gaps may be necessary. For an assessment of the current state-of-the-art in computational linguistics, see Jurafsky & Martin (2009), or Cole et al. (1997). These resources identify the basic computational techniques and machine learning mechanisms that are currently being used. The synthetic teammate research diverges from prevailing computational linguistic approaches in adhering to well-established cognitive constraints on human language processing. Competing computational linguistic approaches make no commitment to cognitive plausibility, relying instead on the use of advanced computational techniques and statistical machine learning methods which are typically not cognitively plausible. Determining how to integrate such techniques to fill gaps in the synthetic teammate research without sacrificing overall cognitive plausibility will be an important research topic. Key components of the NLI toolkit are likely to include: 1) An installation script for rapid installation of the NLI toolkit software, 2) a GUI/programmatic interface that supports the creation of NL interfaces, using synthetic teammate technology in the background, 3) a communications interface for integrating NL interfaces created with the toolkit into training simulations. It is expected that the NL interface will run in a separate process from the training simulation. The NL interface will also need to interface with additional software that performs the task associated with the role of the constructive entity (e.g. pilot, navigator, sensor operator, intelligence analyst), 4) tools for expanding linguistic and world knowledge, and 5) tools to support creation of domain specific knowledge. To demonstrate the capabilities of the NLI toolkit, the NLI toolkit will be used to develop a language capable constructive entity. After development, the constructive entity will be functionally validated in an AF relevant training simulation that involves communication with human teammates. PHASE I: Determine the requirements for creation of a text-chat based NLI toolkit which leverages synthetic teammate technology; identify gaps in the existing technology and propose solutions for closing the gaps; and document the results in a technical report. PHASE II: Research and develop a Natural Language Interface (NLI) toolkit which leverages synthetic teammate technology, identifies gaps in existing technology and provides solutions to fill the gaps. Develop a language capable constructive entity using the NLI toolkit, and demonstrate and validate the use of the constructive entity in an AF-relevant training simulation. Generate a technical report documenting the results. PHASE III: Military applications include creation of language capable constructive entities for training and simulation. Commercial applications include training & simulation, as well as many other applications which require a text-chat based NLI. REFERENCES: 1. Ball, J. (2011). A Pseudo-Deterministic Model of Human Language Processing. In L. Carlson, C. Hlscher, & T. Shipley (Eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society, 495-500. Austin, TX: Cognitive Science Society. 2. Ball, J., Myers, C. W., Heiberg, A., Cooke, N. J., Matessa, M., Freiman, M. & Rodgers, S. (2010). The Synthetic Teammate Project. Computational & Mathematical Organization Theory, 16:3, 271-199. 3. Cole, R., Mariani, J., Uszkoreit, H., Battista Varile, G., Zaenen, A., Zampolli, A. & Zue, V. (1997). Survey of the state of the art in human language technology. NY: Cambridge University Press. 4. Jurafsky, D. & Martin, J. (2009). Speech and Language Processing (second edition). NY: Pearson Prentice Hall. 5. Rodgers, S., Myers, C., Ball, J. & Freiman, M. (2011). The Situation Model in the Synthetic Teammate Project. Proceedings of the 20th Annual Conference on Behavior Representation in Modeling and Simulation, 66-73. Sundance, UT: BRiMS. 6. Rodgers, S., Myers, C., Ball, J. & Freiman, M. (in press). Toward a Situation Model in a Cognitive Architecture. Computational & Mathematical Organization Theory.
US Flag An Official Website of the United States Government