Janet Hitzeman

Centering: A Parametric Theory and Its Instantiations

Centering theory is the best-known framework for theorizing about local coherence and salience; however, its claims are articulated in terms of notions which are only partially specified, such as utterance, realization, or ranking. A great deal of research has attempted to arrive at more detailed specifications of these parameters of the theory; as a result, the claims of centering can be instantiated in many different ways. We investigated in a systematic fashion the effect on the theorys claims of these different ways of setting the parameters. Doing this required, first of all, clarifying what the theorys claims are (one of our conclusions being that what has become known as Constraint 1 is actually a central claim of the theory). Secondly, we had to clearly identify these parametric aspects: For example, we argue that the notion of pronoun used in Rule 1 should be considered a parameter. Thirdly, we had to find appropriate methods for evaluating these claims. We found that while the theorys main claim about salience and pronominalization, Rule 1a preference for pronominalizing the backward-looking center (CB)is verified with most instantiations, Constraint 1a claim about (entity) coherence and CB uniquenessis much more instantiation-dependent: It is not verified if the parameters are instantiated according to very mainstream views (vanilla instantiation), it holds only if indirect realization is allowed, and is violated by between 20 and 25 of utterances in our corpus even with the most favorable instantiations. We also found a trade-off between Rule 1, on the one hand, and Constraint 1 and Rule 2, on the other: Setting the parameters to minimize the violations of local coherence leads to increased violations of salience, and vice versa. Our results suggest that entity coherencecontinuous reference to the same entitiesmust be supplemented at least by an account of relational coherence.

Learning to Resolve Bridging References

We use machine learning techniques to find the best combination of local focus and lexical distance features for identifying the anchor of mereological bridging references. We find that using first mention, utterance distance, and lexical distance computed using either Google or WordNet results in an accuracy significantly higher than obtained in previous experiments.

SpatialML: annotation scheme, resources, and evaluation

SpatialML is an annotation scheme for marking up references to places in natural language. It covers both named and nominal references to places, grounding them where possible with geo-coordinates, and characterizes relationships among places in terms of a region calculus. A freely available annotation editor has been developed for SpatialML, along with several annotated corpora. Inter-annotator agreement on SpatialML extents is 91.3 F-measure on a corpus of SpatialML-annotated ACE documents released by the Linguistic Data Consortium. Disambiguation agreement on geo-coordinates on ACE is 87.93 F-measure. An automatic tagger for SpatialML extents scores 86.9 F on ACE, while a disambiguator scores 93.0 F on it. Results are also presented for two other corpora. In adapting the extent tagger to new domains, merging the training data from the ACE corpus with annotated data in the new domain provides the best performance.

Using Dialogue Features to Predict Trouble During Collaborative Learning

A web-based, collaborative distance-learning system that will allow groups of students to interact with each other remotely and with an intelligent electronic agent that will aid them in their learning has the potential for improving on-line learning. The agent would follow the discussion and interact with the participants when it detects learning trouble of some sort, such as confusion about the problem they are working on or a participant who is dominating the discussion or not interacting with the other participants. In order to recognize problems in the dialogue, we investigated conversational elements that can be utilized as predictors for effective and ineffective interaction between human students. These elements can serve as the basis for student and group models. In this paper, we discuss group interaction during collaborative learning, our representation of participant dialogue, and the statistical models we are using to determine the role being played by a participant at any point in the dialogue and the effectiveness of the group. We also describe student and group models that can be built using conversational elements and discuss one set that we built to illustrate their potential value in collaborative learning.

Towards intelligent agents for collaborative learning: recognizing the roles of dialogue participants

Our goal is to build and evaluate a web-based, collaborative distance-learning system that will allow groups of students to interact with each other remotely and with an intelligent agent that will aid them in their learning. The agent will follow the discussion and interact when it detects learning trouble of some sort, such as confusion about the problem they are working on or a participant who is dominating the discussion or not interacting with the other participants. In order to recognize problems in the dialogue, we are first examining the role that a participant is playing as the dialogue progresses. In this paper we discuss group interaction during collaborative learning, our representation of participant roles, and the statistical model we are using to determine the role being played by a participant at any point in the dialogue.

Text type and the position of a temporal adverbial within the sentence

Consider example (a), below. When the temporal adverbial since 1992 is in sentence-final position as in (i.a), it can attach syntactically at the VP-level or at sentence-level: i. a. Mary has worked in Amsterdam since 1992. b. Since 1992 Mary has worked in Amsterdam. n nHitzeman (1993, 1997) argues that these different positions allow it to take on two readings: one in which there was some period between 1992 and speech time during which Mary worked in Amsterdam and another in which Mary has worked in Amsterdam for the period from 1992 until speech time. In contrast, sentence (i.b), in which the adverbial must attach at sentence-level, has only the second reading. If an initial-position adverbial unambiguously specifies the time of the event expressed by a sentence, then it should be a useful tool for a reader trying to determine the order of events in a narrative. To test the hypothesis that initial-position adverbials occur more often in texts describing events with some temporal order (i.e., a story line), I compare the use of these adverbials in narrative text and in non-narratives. The results show that significantly more initial-position adverbials are used in narratives. I then test the individual narratives and show that the significant difference in use of initial-position adverbials is correlated with the amount of flashback material in a narrative, i.e., with the complexity of the story line.

Maytag: a multi-staged approach to identifying complex events in textual data

We present a novel application of NLP and text mining to the analysis of financial documents. In particular, we describe an implemented prototype, Maytag, which combines information extraction and subject classification tools in an interactive exploratory framework. We present experimental results on their performance, as tailored to the financial domain, and some forward-looking extensions to the approach that enables users to specify classifications on the fly.