Deepak Ramachandran, Mark Fanty, Ronald Provine, Peter Yeh, William Jarrold, Adwait Ratnaparkhi, Benjamin Douglas. Proceedings of the 16th Annual Meeting of the Special Interest Group on Discourse and Dialogue. 2015.

In this paper, we present a speech-driven second screen application for TV program discovery. We give an overview of the application and its architecture. We also present a user study along with a failure analysis. The results from the study are encouraging, and demonstrate our application’s effectiveness in the target domain. We conclude with a discussion of follow-on efforts to further enhance our application.

In this paper, we present an end-to-end dialog system for TV program discovery that uniquely combines several technologies such as trainable relation extraction, belief tracking over relational structures, mixed-initiative dialog management, and inference over large-scale knowledge graphs. We present an evaluation of our end-to-end system with real users, and found that our system performed well along several dimensions such as usability and task success rate. These results demonstrate the effectiveness of our system in the target domain.

In this paper, we present an approach that jointly infers the boundaries of tokens and their labels to construct dictionaries for Information Extraction. Our approach for joint-inference is based on graph propagation, and extends it in two novel ways. First, we extend the graph representation to capture ambiguities that occur during the token extraction phase. Second, we modify the labeling phase (i.e., label propagation) to utilize this new representation, allowing evidence from labeling to be used for token extraction. Our evaluation shows these extensions (and hence our approach) significantly improve the performance of the outcome dictionaries over pipeline-based approaches by preventing aggressive commitment. Our evaluation also shows that our extensions over a base graph-propagation framework improve the precision without hurting the recall.

Poor quality data is a growing and costly problem that af- fects many enterprises across all aspects of their business ranging from operational efficiency to revenue protection. In this paper, we present an application – Data Quality Rules Accelerator (DQRA) – that accelerates Data Quality (DQ) efforts (e.g. data profiling and cleansing) by automatically discovering DQ rules for detecting inconsistencies in data. We then present two evaluations. The first evaluation compares DQRA to existing solutions; and shows that DQRA either outperformed or achieved performance comparable with these solutions on metrics such as precision, recall, and runtime. The second evaluation is a case study where DQRA was piloted at a large utilities company to improve data quality as part of a legacy migration effort. DQRA was able to discover rules that detected data inconsistencies directly impacting revenue and operational efficiency. Moreover, DQRA was able to significantly reduce the amount of effort required to develop these rules compared to the state of the practice. Finally, we describe ongoing efforts to deploy DQRA.

Poor quality data is a growing problem that affects many enterprises across all aspects of their business ranging from operational efficiency to revenue protection. Moreover, this problem is costly to fix because significant effort and resources are required to identify a comprehensive set of rules that can detect (and correct) data defects along various data quality dimensions such as consistency, conformity, and more. Hence, many organizations employ only basic data quality rules that check for null values, format, etc. in efforts such as data profiling and data cleansing; and ignore rules that are needed to detect deeper problems such as inconsistent values across interdependent attributes. This oversight can lead to numerous problems such as inaccurate reporting of key metrics used to inform critical decisions or derive business insights. In this paper, we present an approach that efficiently and robustly discovers data quality rules - in particular conditional functional dependencies - for detecting inconsistencies in data and hence improves data quality along the critical dimension of consistency. We evaluate our approach empirically on several real-world data sets. We show that our approach performs well on these data sets for metrics such as precision and recall. We also compare our approach to an established solution and show that our approach outperforms this solution for the same metrics. Finally, we show that our approach scales efficiently with the number of records, the number of attributes, and the domain size.

We describe a problem solver built to answer questions like those on advanced placement exams using knowledge bases authored by domain experts. The problem solver is designed to work independently of any particular knowledge base or domain. Given a question, the problem solver identifies those portions of the knowledge base that are relevant to the question. We found that simple heuristics for judging relevance significantly improved performance, with no drop in coverage.

In this paper, we give a progress report on an ongoing effort at Accenture to develop a technology platform for building a wide range of corporate radar applications,which can turn the Web into a systematic source of business insight. Our goal is to share the platform we have developed and the lessons we have learned, so others can leverage this knowledge when building similar applications. We give an overview of this platform, which integrates a combination of established AI technologies - i.e. semantic models, natural language processing, and inference engines - in a novel way. We then illustrate the kinds of corporate radars that can be built with our platform through two applications we developed at Accenture: the technology lifecycle tracker, which assesses the maturity of technologies from the wireless industry, and the technology trend tracker, which measures hype versus reality for emerging technology trends such as cloud computing, software-as-a-service, and more. Finally, we discuss our experiences in using this platform to build these applications and the lessons learned.

In this paper, we present a knowledge based approach to capture rich semantic representations from online news sources for business intelligence (BI) applications that know the representations of interest in advance. Our approach performs this task by generating phrases from these representations and matching these phrases against the news using a set of syntactic and semantic transformations. The representation that best matches a piece of news is selected as its meaning. We present two evaluations showing how our approach performs well on capturing the semantics of online new sources for BI applications.

As part of the ongoing project, Project Halo, our goal is to build a system capable of answering questions posed by novice users to a formal knowledge base. In our current context, the knowledge base covers selected topics in physics, chemistry, and biology, and our question set consists of AP (advanced high-school) level examination questions. The task is challenging because the questions are linguistically complex and are often incomplete (assume unstated knowledge), and because the users do not have prior knowledge of the system's contents. Our solution involves two parts: a controlled language interface, in which users reformulate the original natural language questions in a simplified version of English, and a novel problem solver that can elaborate initially inadequate logical interpretations of a question by selecting relevant pieces of knowledge in the knowledge base. An evaluation of the work in 2006 showed that this approach is feasible and that complex, multisentence questions can be posed and answered, thus illustrating novel ways of dealing with the knowledge capture impedance between users and a formal knowledge base, while also revealing challenges that still remain.

An important problem in artificial intelligence is capturing, from natural language, formal representationsallthat can be used by a reasoner to compute an answer. Many researchers have studied this problem by developing algorithms addressing specific phenomena in natural language interpretation, but few have studied (or cataloged) the types of failures associated with this problem. Knowledgeallof these failures can help researchers by providing a roadallmap of open research problems and help practitioners by providing a checklist of issues to address in order to build systems that can achieve good performance on this problem.allIn this paper, we present a study -- conducted in the context of the Halo Project -- cataloging the types of failures that occur when capturing knowledge from naturallanguage. We identified the categories of failures by examining a corpus of questions posed byallnaive usersallto a knowledge based question answering system and empirically demonstrated the generality of ourallcategorizations. We also describe available technologies that can address some of the failures we have identified.

An ultimate goal of AI is to build end-to-end systems that interpret natural language, reason over the resulting logical forms, and perform actions based on that reasoning. This requires systems from separate fields be brought together, but often this exposes representational gaps between them. The logical forms from a language interpreter may mirror the surface forms of utterances too closely to be usable as-is, given a reasoner's requirements for knowledge representations. What is needed is a system that can match logical forms to background knowledge flexibly to acquire a rich semantic model of the speaker's goal. In this paper, we present such a "matcher" that uses semantic transformations to overcome structural differences between the two representations. We evaluate this matcher in a MUC-like template-filling task and compare its performance to that of two similar systems.

Many AI tasks require determining whether two knowledge representations encode the same knowledge. Solving this matching problem is hard because representations may encode the same content but differ substantially in form. Previous approaches to this problem have used either syntactic measures, such as graph edit distance, or semantic knowledge to determine the "distance" between two representations. Although semantic approaches outperform syntactic ones, previous research has focused primarily on the use of taxonomic knowledge. We show that this is not enough because mismatches between representations go largely unaddressed. In this paper, we describe how transformations can augment existing semantic approaches to further improve matching. We also describe the application of our approach to the task of critiquing military Courses of Action and compare its performance to other leading algorithms.