Manovaad: A Novel Approach to Event Oriented Corpus Creation Capturing Subjectivity and Focus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 476–484 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 476 The Margarita Dialogue Corpus: A Data Set for Time-Offset Interactions and Unstructured Dialogue Systems Alberto M. Chierici, Nizar Habash, Margarita Bicec Computational Approaches to Modeling Language (CAMeL) Lab New York University Abu Dhabi {alberto.chierici, nizar.habash, margarita.bee}@nyu.edu Abstract Time-Offset Interaction Applications (TOIAs) are systems that simulate face-to-face conversations between humans and digital human avatars recorded in the past. Developing a well-functioning TOIA involves several research areas: artificial intelligence, human-computer interaction, natural language processing, question answering, and dialogue systems. The first challenges are to define a sensible methodology for data collection and to create useful data sets for training the system to retrieve the best answer to a user’s question. In this paper, we present three main contributions: a methodology for creating the knowledge base for a TOIA, a dialogue corpus, and baselines for single-turn answer retrieval. We develop the methodology using a two-step strategy. First, we let the avatar maker list pairs by intuition, guessing what possible questions a user may ask to the avatar. Second, we record actual dialogues between random individuals and the avatar-maker. We make the Margarita Dialogue Corpus available to the research community. This corpus comprises the knowledge base in text format, the video clips for each answer, and the annotated dialogues. Keywords: Corpus Creation, Annotation, Dialogue, Question Answering 1. Introduction Time-Offset Interaction Applications (TOIAs) are specific types of question answering systems that simulate face-to- face conversations between humans (interactors or inter- rogators) and previously-recorded digital human avatars which were created by avatar makers. Such conversa- tional avatars are important for preserving personal and cross-generational histories, as well as, for teaching and coaching, among other possible applications. Our work on TOIAs is heavily inspired by Artstein et al. (2015), who first introduced the term time-offset interaction. Developing, streamlining, and making the process of cre- ating TOIAs affordable are important goals. But building such systems is not an easy task because the interaction with the avatars should ideally be as close as possible to a real, human-to-human interaction. This critical feature poses challenges from an engineering point of view (for instance, connecting the video clips flawlessly) and from many other points of view related to dialogue management and natural language processing. We refer to the work by Abu Ali et al. (2018) for a detailed description of the TOIA platform we use for storing and accessing video clips, and for interpreting the interrogator’s input. The first step to build a TOIA involves the creation of its Knowledge Base. This database consists of a set of pairs, and answer video clips. The answer clips include record- ings of the avatar’s answers, and clarification requests (e.g., “Can you repeat your question?”), as well as filler se- quences to play while the system is waiting for the next question (e.g., the avatar adjusting her hair). One interest- ing problem is how to best …

Content truncated to 500 words