We investigate coreference relationships between NPs with the same head noun. It is relatively common in unsupervised work to assume that such pairs are coreferent { but this is not always true, especially if realistic mention detection is used. We describe the distribution of non-coreferent same-head pairs in news and conversational discourse. We present an unsupervised generative model which learns not to link some same-head NPs using syntactic features. Our model improves precision on both news and conversation. 1.

syntactic chunking with acoustic cues:

A well-known effect in speech production is that more predictable words tend to be phonetically reduced. Recent work has suggested that predictability effects result from hardwired properties of the language production system, rather than active modulation by the talker to accommodate the listener. However, these studies investigated only minor manipulations of listener characteristics. Here, we examine predictability effects with two very different listener populations: adults and preverbal infants. Using mixed effects regressions on spontaneous speech corpora, we compare the effect of word frequency, probability in context, and previous mention on word duration in adult-directed and infant-directed speech. We find that the effects of preceding context and word frequency differ according to listener. Contrary to previous work, these results suggest that talkers do modulate the phonetic effects of predictability based on listener characteristics. To our knowledge, this study is also the first published analysis of predictability effects in infant-directed speech.

Cue words ◮ Cue words are short responses which indicate how discourse structures are to be updated with respect to a new utterance.

We discuss the collection and analysis of a cross-sectional and longitudinal learner corpus consisting of answers to reading comprehension questions written by adult second language learners of German. We motivate the need for such task-based learner corpora and identify the properties which make reading comprehension exercises a particularly interesting task. In terms of the creation of the corpus, we introduce the web-based WELCOME tool we developed to support the decentralized data collection and annotation of the richly structured corpus in real-life language teaching programs. On the analysis side, we investigate the binary and the complex content-assessment classification scheme used by the annotators and the inter-annotator agreement obtained for the current corpus snapshot, at the halfway point of our four-year effort. We present results showing that for such task-based corpora, meaning assessment can be performed with reasonable agreement and we discuss several sources of disagreement. 1

While information status (IS) plays a crucial role in discourse processing, there have only been a handful of attempts to automatically determine the IS of discourse entities. We examine a related but more challenging task, fine-grained IS determination, which involves classifying a discourse entity as one of 16 IS subtypes. We investigate the use of rich knowledge sources for this task in combination with a rule-based approach and a learning-based approach. In experiments with a set of Switchboard dialogues, the learning-based approach achieves an accuracy of 78.7%, outperforming the rulebased approach by 21.3%. 1

Abstract not found