Conversational agent for aerospace question answering: A position paper

Gérard M Dupont

Abstract: This paper introduces a research problem currently faced in the use of conversational agents within the aerospace industry. The aerospace domain is characterized by products and systems that are built over decades of engineering to reach high levels of performance within complex environments. This has resulted in technical documentation piling up over time. The natural next step is to optimize the access to such document repositories to ensure completeness and relevance of information requested by operators. Recent developments in natural language processing, text understanding and information retrieval have provided major breakthroughs in the form of conversational interfaces for contextual search and cognitive assistants. However, the context of aerospace product documentation raises a number of specific challenges that are currently not fully addressed by advances in conversational agents intended for the general public. This position paper describes some of these particularities. Even if these are not unique for the domain nor new, they serve to assess the recent progress in evaluation protocols and initiatives for conversational agents. Finally, we propose an evaluation protocol that may address these challenges.