Disambiguation and Filtering Methods in Using Web Knowledge for Coreference Resolution

The authors investigate two publicly available Web knowledge bases, Wikipedia and Yago, in an attempt to leverage semantic information and increase the performance level of a state-of-the-art coreference resolution engine. They extract semantic compatibility and aliasing information from Wikipedia and Yago, and incorporate it into a coreference resolution system. The authors show that using such knowledge with no disambiguation and filtering does not bring any improvement over the baseline, mirroring the previous findings (Ponzetto & Poesio, 2009). They propose, therefore, a number of solutions to reduce the amount of noise coming from Web resources: using disambiguation tools for Wikipedia, pruning Yago to eliminate the most generic categories and imposing additional constraints on affected mentions. The evaluation experiments on the ACE-02 corpus show that the knowledge, extracted from Wikipedia and Yago, improves the system’s performance by 2-3 percentage points.

Guru

The authors discuss Guru, a conversational expert ITS. Guru is designed to mimic expert human tutors using advanced applied natural language processing techniques including natural language understanding, knowledge representation, and natural language generation.

Understanding and Reasoning with Text

Consider the assignment that teachers have been giving their students for years: “Write an expository essay on a scientific topic. Example topics may include global warming, human memory, or the spread of infectious diseases. You must have at least three references.” The instructor makes it clear that the paper should have a thesis or claim that is supported by evidence. Claims might be that global warming will be disastrous only for some nations, why it is futile to teach mnemonics to young children, or that cell phone use causes cancer. From the perspective of the student (and cognitive psychologists), this assignment is challenging at any grade. The challenge is that the assignment entails a number of complicated and interconnected tasks. For example, reading a research paper requires the reader to make inferences that span sentences and paragraphs (in addition to a whole host of other processes), and to understand the logical and rhetorical structure of the text as a whole. If the paper describes an experiment, the student must additionally understand how to determine whether the data support the conclusion (i.e., the scientific method). In most cases, the student must also integrate the content of several papers (sources) into a coherent structure. This process involves evaluating the credibility of the sources, selecting relevant pieces of information from each, and putting them into a coherent argument structure. No wonder such assignments are met with groans.

Cognitive Load Aspects of Text Processing

Cognitive load theory investigates instructional consequences of processing limitations of the human cognitive system. Because of these limitations, text processing may result in an excessive cognitive load that would influence comprehension and learning from texts, as well as change learner affective states. This chapter reviews basic assumptions of cognitive load theory, their consequences for optimizing the design of information presentations, and implications for processing written and spoken texts.

Extracting Commonsense Knowledge Using Concepts Properties

Commonsense knowledge encompasses facts that people know but do not communicate most of the time. For example, one needs water and soap to take a shower is commonsense. This chapter presents a semantically grounded method for extracting commonsense knowledge. First, commonsense rules are identified, e.g., one cannot see imaginary objects. Second, those rules are combined with a basic semantic representation in order to infer commonsense facts, e.g. one cannot see a flying carpet. Further combinations of semantic relations with inferred commonsense facts are proposed and analyzed. Experimental results show that this novel method is able to extract thousands of commonsense facts with little human interaction and high accuracy.

Multiple Influences on the Use of English Spatial Prepositions

What do people know when they know a word? Previous accounts of the semantics of spatial locatives suggest that spatial meaning is based on both geometric and extra-geometric aspects of spatial scenes. However, attempts to explicitly delineate different sources of extra-geometric influences are still comparatively rare; even more rare are attempts to combine these different sources so as to examine their interactions. This chapter presents four studies examining the ways in which three classes of attributes – geometric, functional, and qualitative physical – influence speakers’ uses of the English spatial prepositions in and on. The experiments show that all three kinds of factors play roles in English speakers’ choice between these prepositions. The chapter concludes that the semantics of spatial locatives must take into account a complex set of interacting factors.

Text-Based Affect Detection in Intelligent Tutors

Affect-sensitive Intelligent Tutoring Systems are an exciting new educational technology that aspire to heighten motivation and enhance learning gains in interventions that are dynamically adaptive to learners’ affective and cognitive states. Although state of the art affect detection systems rely on behavioral and physiological measures for affect detection, we show that a textual analysis of the tutorial discourse provides important cues into learners’ affective states. This chapter surveys the existing literature on text-based affect sensing and focuses on how learners’ affective states (boredom, flow/engagement, confusion, and frustration) can be automatically predicted by variations in the cohesiveness of tutorial dialogues during interactions with AutoTutor, an intelligent tutoring system with conversational dialogues. The authors discuss the generalizability of findings to other domains and tutoring systems, the possibility of constructing real-time cohesion-based affect detectors, and implications for text-based affect detection for the next generation affect-sensitive learning environments.

Fairy Tales and ESL Texts

This chapter describes a study that investigates the potential value of using traditional fairy tales as reading material for English language learners (ELL). Using the computational textual analysis software, the Gramulator, the authors analyzed the linguistic features of fairy tales relative to a corpus of ELL reading material and a corpus of baseline educational texts for native English speakers. The results of the analyses suggest that there are significant similarities between fairy tales and ESL texts, but differences lie in the content of the text types, with fairy tales appearing significantly more narrative in style and ESL texts appearing more expository. The study has important implications for educators and materials developers in the field of English as a Second Language.

Newness and Givenness of Information

The identification of new versus given information within a text has been frequently investigated by researchers of language and discourse. Despite theoretical advances, an accurate computational method for assessing the degree to which a text contains new versus given information has not previously been implemented. This study discusses a variety of computational new/given systems and analyzes four typical expository and narrative texts against a widely accepted theory of new/given proposed by Prince (1981). Findings suggest that a latent semantic analysis (LSA) based measure called span outperforms standard LSA in detecting both new and given information in text. Further, the span measure outperforms standard LSA for distinguishing low versus high cohesion versions of text. Results suggest that span may be a useful variable in a wide array of discourse analyses.

Information Extraction from Text and Beyond

In this chapter, the author defines information extraction from text, describes common information extraction tasks, and discusses current information extraction issues being the need to develop technologies that require a minimum of human supervision, to build systems that automatically acquire world knowledge, and to integrate their outputs into advanced information extraction systems. Current emerging research on extraction of narrative scenarios and complex concepts revives an old dream and opens a way to full natural language understanding.