Semantic Grounding in Dialogue for Complex Problem Solving

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 841–850, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Semantic Grounding in Dialogue for Complex Problem Solving Xiaolong Li Kristy Elizabeth Boyer Department of Computer Science North Carolina State University Department of Computer Science North Carolina State University Raleigh, NC, 27695 Raleigh, NC, 27695 xli30@ncsu.edu keboyer@ncsu.edu Abstract Dialogue systems that support users in complex problem solving must interpret user utterances within the context of a dy- namically changing, user-created problem solving artifact. This paper presents a novel approach to semantic grounding of noun phrases within tutorial dialogue for com- puter programming. Our approach per- forms joint segmentation and labeling of the noun phrases to link them to attributes of entities within the problem-solving envi- ronment. Evaluation results on a corpus of tutorial dialogue for Java programming demonstrate that a Conditional Random Field model performs well, achieving an accuracy of 89.3% for linking semantic segments to the correct entity attributes. This work is a step toward enabling dia- logue systems to support users in increas- ingly complex problem-solving tasks. 1 Introduction In the dialogue systems research community, there is growing recognition that dialogue systems need to support users in increasingly complex tasks. To move in this direction, dialogue systems must per- form natural language understanding within richer and richer contexts, and this understanding in- cludes semantic interpretation of user utterances (Traum, et al., 2012, Rudnicky, et al., 1999). Pre- vious approaches for semantic interpretation in- clude domain-specific grammars (Lemon et al., 2001) and open-domain parsers together with a domain-specific lexicon (Rosé, 2000). However, existing techniques are not sufficient to support increasingly complex problem-solving dialogues due to several challenges. For example, domain- specific grammars become intractable when ap- plied to more ill-formed domains, and open- domain parsers may not perform well across do- mains (McClosky et al., 2010). The call for addressing these limitations is par- ticularly strong for dialogue systems that help peo- ple learn, such as tutorial dialogue systems. Today’s tutorial dialogue systems engage in natu- ral language dialogue in support of tasks such as solving qualitative physics problems (VanLehn et al., 2002), understanding computer architecture and physics (Graesser et al., 2004), and predicting behavior of electrical circuits (Dzikovska et al., 2011). Although these systems differ in many ways, they have an important commonality: in or- der to semantically interpret user dialogue utter- ances, these systems ground the utterances in a fixed domain description that is an integral part of the engineered system. This characteristic is shared by most dialogue systems, which ground their dia- logue in manually defined domain-specific ontolo- gies, such as for the task of booking flights (Allen, et al., 2001), checking bus schedules (Raux, 2004), and finding restaurants (Young et al., 2007). These task-oriented domains, though they pre- sent a rich set of research challenges, stand in stark contrast to a complex problem-solving domain in which the user is creating an artifact to solve a …

Content truncated to 500 words