I will first give an idiosyncratic overview of NLP in the UK, based on my visits to York, Oxford, Sheffield, Aberdeen, the Open U, Cambridge, Ulster, and Dublin City Universities during my sabbatical.
I will also briefly describe my joint research with Johanna Moore's group at the University of Edinburgh, which investigates whether previous findings and methods in the area of tutorial dialogue can be generalized across dialogue corpora that differ in domain (physics versus electricity), modality (spoken versus typed), and tutor type (human versus computer)]]>
Determining whether a condition is historical or recent is important for accurate results in biomedicine. In this paper, we investigate four types of information found in clinical text that might be used to make this distinction. We conducted a descriptive, exploratory study using annotation on clinical reports to determine whether this temporal information is useful for classifying conditions as historical or recent. Our initial results suggest that few of these feature values can be used to predict temporal classification. ]]>
This paper introduces ONYX, a sentence-level text analyzer that
implements a number of innovative ideas in syntactic and semantic analysis.
ONYX is being developed as part of a project that seeks to translate spoken
dental examinations directly into chartable findings.
ONYX integrates syntax
and semantics to a high degree. It interprets sentences using a combination
of probabilistic classifiers, graphical unification, and semantically
annotated grammar rules. In this preliminary evaluation, ONYX shows
inter-annotator agreement scores with humans of 86% for assigning semantic
types to relevant words, 80% for inferring relevant concepts from words, and
76% for identifying relations between concepts.]]>
I will first give an idiosyncratic overview of NLP in the UK, based on my visits to York, Oxford, Sheffield, Aberdeen, the Open U, Cambridge, Ulster, and Dublin City Universities during my sabbatical.
I will also briefly describe my joint research with Johanna Moore's group at the University of Edinburgh, which investigates whether previous findings and methods in the area of tutorial dialogue can be generalized across dialogue corpora that differ in domain (physics versus electricity), modality (spoken versus typed), and tutor type (human versus computer).]]>
TBA ]]>
In my thesis study, I investigate how to evaluate and how to build user simulations to help dialog system development.When evaluating user simulations, I use both human judges and automatic evaluation measures to assess the simulation model qualities. When building user simulations, I examine three factors that impact simulation models in the tasks of dialog strategy learning and dialog system development. (This is a practice talk for Hua's postdoc interview.)]]>
The second study is based on the author's chapter on intertextuality in "Rhetoric Online." The study defines intertextuality as a form of interreference among texts in which an already familiar text is invoked or played upon in a new textual context. Intertextuality makes use of resources in the larger intertext (the textual context in which the discourse appears) to involve users in active co-construction of the text¹s meaning. A number of contrasting examples of this phenomenon will be introduced and discussed, so that the kinds of variation in online intertextuality can be identified.]]>
Opinion analysis focuses on extracting subjective expressions and sentiments from text. Large-scale availability of public opinion over the internet weblogs, review websites, discussion forums and face to face conversations has motivated a great amount of research in this area in recent times.
Much of the research in opinion analysis focuses on finding expressions that are opinion-indicators (for example, expressions like ³great², ³boring², or ³failed to kill the spirit²). We feel that such approaches can be augmented by discourse-level analysis of opinions. Discourse-level analysis allows opinion expressions in different parts of the discourse to be interpreted in an interdependent, coherent fashion and allows us to capture the varied ways in which people reveal their opinions.
In order to achieve discourse-level opinion analysis, we first develop a scheme to relate different opinion expressions in a discourse. We validate this design by showing that human annotators can reliably recognize the opinion relations. Then, we implement our scheme with a machine learning framework of collective inference to show that a computer¹s ability to recognize opinion expressions is indeed enhanced when the discourse level information is used. Finally, we show that, by performing feature engineering using a simple machine learning framework, the discourse-level opinion relations of our scheme can be automatically detected.]]>
Active learning has proven to be a successful strategy in quick
development of corpora to be used in training of statistical natural
language parsers.
A vast majority of studies in this field has focused on
estimating informativeness of samples; however, representativeness of
samples is another important criterion to be considered in active learning.
We present a novel metric for estimating representativeness of sentences, based on a modification of Zipf's Principle of Least Effort. Experiments on WSJ corpus with a wide-coverage parser show that our method performs always at least as good as and generally significantly better than alternative representativeness-based methods.]]>
This paper analyzes opinion categories like Sentiment and Arguing in meetings.
We first annotate the categories manually. We then develop genre-specific lexicons using interesting function word combinations for detecting the opinions. We analyze relations between dialog structure information and opinion expression in context of multi-party discourse. Finally we show that classifiers using lexical and discourse knowledge have significant improvement over baseline.