Michelle L. Gregory

Effects of disfluencies, predictability, and utterance position on word form variation in English conversation

Function words, especially frequently occurring ones such as (the, that, and, and of), vary widely in pronunciation. Understanding this variation is essential both for cognitive modeling of lexical production and for computer speech recognition and synthesis. This study investigates which factors affect the forms of function words, especially whether they have a fuller pronunciation (e.g., thi, thaet, aend, inverted-v v) or a more reduced or lenited pronunciation (e.g., thax, thixt, n, ax). It is based on over 8000 occurrences of the ten most frequent English function words in a 4-h sample from conversations from the Switchboard corpus. Ordinary linear and logistic regression models were used to examine variation in the length of the words, in the form of their vowel (basic, full, or reduced), and whether final obstruents were present or not. For all these measures, after controlling for segmental context, rate of speech, and other important factors, there are strong independent effects that made high-frequency monosyllabic function words more likely to be longer or have a fuller form (1) when neighboring disfluencies (such as filled pauses uh and um) indicate that the speaker was encountering problems in planning the utterance; (2) when the word is unexpected, i.e., less predictable in context; (3) when the word is either utterance initial or utterance final. Looking at the phenomenon in a different way, frequent function words are more likely to be shorter and to have less-full forms in fluent speech, in predictable positions or multiword collocations, and utterance internally. Also considered are other factors such as sex (women are more likely to use fuller forms, even after controlling for rate of speech, for example), and some of the differences among the ten function words in their response to the factors.

User-directed Sentiment Analysis: Visualizing the Affective Content of Documents

Recent advances in text analysis have led to finer-grained semantic analysis, including automatic sentiment analysis---the task of measuring documents, or chunks of text, based on emotive categories, such as positive or negative. However, considerably less progress has been made on efficient ways of exploring these measurements. This paper discusses approaches for visualizing the affective content of documents and describes an interactive capability for exploring emotion in a large document collection.

Topicalization and left-dislocation: a functional opposition revisited☆

Abstract In this case study, we use conversational data from the Switchboard corpus to investigate the functional opposition between two pragmatically specialized constructions of English: Topicalization and Left-Dislocation. Specifically, we use distribution trends in the Switchboard corpus to revise several conclusions reached by Prince (1981a,b, 1997) concerning the function of Left-Dislocation. While Prince holds that Left-Dislocation has no unitary functionn, we argue that the distinct uses of the construction identified by Prince can be subsumed under the general function of topic promotion. While Prince holds that Topicalization is a more pragmatically specialized construction than Left-Dislocation, we argue that Left-Dislocation has equally restrictive and distinct use conditions, which reflect its status as a topic-promoting device. We conclude that computational corpus methods provide an important check on the validity of claims concerning pragmatic markedness.

PNNL: A Supervised Maximum Entropy Approach to Word Sense Disambiguation

In this paper, we described the PNNL Word Sense Disambiguation system as applied to the English all-word task in SemEval 2007. We use a supervised learning approach, employing a large number of features and using Information Gain for dimension reduction. The rich feature set combined with a Maximum Entropy classifier produces results that are significantly better than baseline and are the highest F-score for the fined-grained English all-words subtask of SemEval.

Describing story evolution from dynamic information streams

Sources of streaming information, such as news syndicates, publish information continuously. Information portals and news aggregators list the latest information from around the world enabling information consumers to easily identify events in the past 24 hours. The volume and velocity of these streams causes information from prior days to quickly vanish despite its utility in providing an informative context for interpreting new information. Few capabilities exist to support an individual attempting to identify or understand trends and changes from streaming information over time. The burden of retaining prior information and integrating with the new is left to the skills, determination, and discipline of each individual. In this paper we present a visual analytics system for linking essential content from information streams over time into dynamic stories that develop and change over multiple days. We describe particular challenges to the analysis of streaming information and present a fundamental visual representation for showing story change and evolution over time.

The effect of language model probability on pronunciation reduction

We investigate how the probability of a word affects its pronunciation. We examined 5618 tokens of the 10 most frequent (function) words in Switchboard and 2042 tokens of content words whose lexical form ends in a t or d. Our observations were drawn from the phonetically hand-transcribed subset of the Switchboard corpus, enabling us to code each word with its pronunciation and duration. Using linear and logistic regression to control for contextual factors, we show that words which have a high unigram, bigram, or reverse bigram (given the following word) probability are shorter, more likely to have a reduced vowel, and more likely to have a deleted final t or d. These results suggest that pronunciation models in speech recognition and synthesis should take into account word probability given both the previous and following words, for both content and function words.

Combining Hierarchical and Associative Gene Ontology Relations With Textual Evidence in Estimating Gene and Gene Product Similarity

Two approaches have recently emerged where the similarity between two genes or gene products is obtained by comparing Gene Ontology (GO) annotations associated with the genes or gene products. One approach captures GO-based similarity in terms of hierarchical relations within each gene subontology, while the other relies on associative relations across the three gene subontologies. We propose a novel methodology where the two approaches can be merged and enriched by textual evidence extracted from biomedical literature with ensuing benefits in coverage and stronger correlation with sequence-based similarity

Understanding the dynamics of collaborative multi-party discourse

In this paper, we discuss the efforts underway at the Pacific Northwest National Laboratory in understanding the dynamics of multi-party discourse across a number of communication modalities, such as email, instant messaging traffic and meeting data. Two prototype systems are discussed. The Conversation Analysis Tool (ChAT) is an experimental test-bed for the development of computational linguistic components and enables users to easily identify topics or persons of interest within multi-party conversations, including who talked to whom, when, the entities that were discussed, etc. The Retrospective Analysis of Communication Events (RACE) prototype, leveraging many of the ChAT components, is an application built specifically for knowledge workers and focuses on merging different types of communication data so that the underlying message can be discovered in an efficient, timely fashion.

Quantifying sentiment and influence in blogspaces

The weblog, or blog, has become a popular form of social media, through which authors can write posts, which can in turn generate feedback in the form of user comments. When considered in totality, a collection of blogs can thus be viewed as a sort of informal collection of mass sentiment and opinion. An obvious topic of interest might be to mine this collection to obtain some gauge of public sentiment over the wide variety of topics contained therein. However, the sheer size of the so-called blogosphere, combined with the fact that the subjects of posts can vary over a practically limitless number of topics poses some serious challenges when any meaningful analysis is attempted. Namely, the fact that largely anyone with access to the Internet can author their own blog, raises the serious issue of credibility---should some blogs be considered to be more influential than others, and consequently, when gauging sentiment with respect to a topic, should some blogs be weighted more heavily than others? In addition, as new posts and comments can be made on almost a constant basis, any blog analysis algorithm must be able to handle such updates efficiently. In this paper, we give a formalization of the blog model. We give formal methods of quantifying sentiment and influence with respect to a hierarchy of topics, with the specific aim of facilitating the computation of a per-topic, influence-weighted sentiment measure. Finally, as efficiency is a specific endgoal, we give upper bounds on the time required to update these values with new posts, showing that our analysis and algorithms are scalable.

Social media and social reality

Social Media provide an exciting and novel view into social phenomena. The vast amounts of data that can be gathered from the Internet coupled with massively parallel supercomputers such as the Cray XMT open new vistas for research. Conclusions drawn from such analysis must recognize that social media are distinct from the underlying social reality. Rigorous validation is essential. This paper briefly presents results obtained from computational analysis of social media - utilizing both blog and twitter data. Validation of these results is discussed in the context of a framework of established methodologies from the social sciences. Finally, an outline for a set of supporting studies is proposed.