Lightly Supervised Learning of Procedural Dialog Systems

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1669–1679, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Lightly Supervised Learning of Procedural Dialog Systems Svitlana Volkova CLSP Johns Hopkins University Baltimore, MD svitlana@jhu.edu Pallavi Choudhury, Chris Quirk, Bill Dolan NLP Group Microsoft Research Redmond, WA pallavic,chrisq, billdol@microsoft.com Luke Zettlemoyer Computer Science and Engineering University of Washington Seattle, WA lsz@cs.washington.edu Abstract Procedural dialog systems can help users achieve a wide range of goals. However, such systems are challenging to build, currently requiring manual engineering of substantial domain-specific task knowl- edge and dialog management strategies. In this paper, we demonstrate that it is pos- sible to learn procedural dialog systems given only light supervision, of the type that can be provided by non-experts. We consider domains where the required task knowledge exists in textual form (e.g., in- structional web pages) and where system builders have access to statements of user intent (e.g., search query logs or dialog interactions). To learn from such tex- tual resources, we describe a novel ap- proach that first automatically extracts task knowledge from instructions, then learns a dialog manager over this task knowledge to provide assistance. Evaluation in a Mi- crosoft Office domain shows that the indi- vidual components are highly accurate and can be integrated into a dialog system that provides effective help to users. 1 Introduction Procedural dialog systems aim to assist users with a wide range of goals. For example, they can guide visitors through a museum (Traum et al., 2012; Aggarwal et al., 2012), teach students physics (Steinhauser et al., 2011; Dzikovska et al., 2011), or enable interaction with a health care U: “I want to add page numbers and a title” S: “Top or Bottom of the page?” U: “Top” S: “Please select page design from the tem- plates” (*System shows drop down menu*) U: *User selects from menu* S: “Enter header or footer content” U: “C.V.” S: “Task completed.” Figure 1: An example dialog interaction between a system (S) and user (U) that can be automatically achieved by learning from instructional web page and query click logs. system (Morbini et al., 2012; Rizzo et al., 2011). However, such systems are challenging to build, currently requiring expensive, expert engineering of significant domain-specific task knowledge and dialog management strategies. In this paper, we present a new approach for learning procedural dialog systems from task- oriented textual resources in combination with light, non-expert supervision. Specifically, we as- sume access to task knowledge in textual form (e.g., instructional web pages) and examples of user intent statements (e.g., search query logs or dialog interactions). Such instructional resources are available in many domains, ranging from recipes that describe how to cook meals to soft- ware help web pages that describe how to achieve goals by interacting with a user interface.1 1ehow.com,wikianswers.com 1669 There are two key challenges: we must (1) learn to convert the textual knowledge into a us- able form and (2) learn a dialog manager that pro- vides …

Content truncated to 500 words