The Language of Learning
Abstract
Approved for Public ReleaseUnder ONR funding, the Language-Endowed Intelligent Agents (LEIA) Lab at Rensselaer Polytechnic Institute has been working on meaning-oriented, domain-independent language processing for cognitive agents. We will advance this program ofwork through a project that focuses on the language of learning. People teach and learn largely through language # more specifically, through the meaning conveyed by language. So, too, must agents who serve as collaborators, assistants, teachers, and advisors. Like all LEIA modeling, this project will pursue two thrusts of R&D simultaneously, domain-independent and domain-specific. Domain-independent thrust: We will develop LEIAs that communicate and reason about teaching, learning, and advising in principle, as applicable to any domain. This involves the computational cognitive modeling of the content, structure, and flow of associated texts and dialogs. All LEIAs operating in teaching/learning/advising environments need certain basic capabilities, such as using their stored knowledge and expectations to fill in the blanks of everyday English, which is full of ellipsis, implicatures, fragmentary utterances and the like. Other capabilities are specific to LEIAs carrying out particular roles. For example, LEIAs who are tasked with learning something from scratch must be prepared to learn new vocabulary and concepts, and to ask for clarification when what the human teacher presents is vague or underspecified; LEIAs who are being interactively instructed to improve their performance on a known task must be prepared to modify their knowledge, workflow, etc.; and LEIAs serving as teachers and advisors must be able to identify fail points in human performance and explain, with appropriate detail, how to improve performance. Domain-specific thrust: We will apply the abovementioned generic capabilities to domains of interest to ONR in collaboration with researchers at NRL. The NRL team exploresvarious aspects of human behavior using simulation environments. Since LEIA researchers work toward operationalizing human behaviorin cognitive agents, this promises to be a fruitful collaboration. The first stage of this project will involve exploring the various simulation environments available to the NRL team and identifying which environments and agent capabilities have the most potential to have real impact for ONR objectives. Two examples of simulation environments that we have discussed are SCOUT (a multi-task environment involving unmanned aerial vehicles) and X-Plane (a flight simulator). These simulators target the domains of supervisory control and pilot training, respectively. They offer a host of interesting opportunities for the computational cognitive modeling of communication, collaboration, learning, teaching, mentoring, and automatic assessment. To give just two examples: SCOUT offers the opportunity to integrate eye tracking results with language understanding, which is a core example of multi-modal perception processing (something that the LEIA#s cognitive architecture accommodates); and X-Plane involves pilot-to-tower communications, which are sufficiently structured in content and form to lend themselves to formal modeling and agent simulation. To summarize, domain-specific development addresses ONR objectives with the goal of building useful agent systems in the near term, whereas domain-independent development endows intelligent agents with core capabilities that can be applied across domains and will contribute to the ultimate goal of general artificial intelligence.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2023
- Source ID
- N000142312060
Entities
People
- Marjorie Joan McShane
Organizations
- Office of Naval Research
- Rensselaer Polytechnic Institute
- United States Navy