Emina Kurtic

Resources for turn competition in overlapping talk

Highlights? We derive decision trees that discriminate turn-competitive and non-competitive overlapping talk. ? F0 and intensity were found to be the most important prosodic resources for turn competition. ? Competitive and non-competitive overlaps could be initiated at a range of different places in the current speakers turn. ? Overlap placement features played a greater role than prosodic features in indicating turn competition. Overlapping talk occurs frequently in multi-party conversations, and is a domain in which speakers may pursue various communicative goals. The current study focuses on turn competition. Specifically, we seek to identify the phonetic differences that discriminate turn-competitive from non-competitive overlaps. Conversation analysis techniques were used to identify competitive and non-competitive overlaps in a corpus of multi-party recordings. We then generated a set of potentially predictive features relating to prosody (F0, intensity, speech rate, pausing) and overlap placement (overlap duration, point of overlap onset, recycling etc.). Decision tree classifiers were trained on the features and tested on a classification task, in order to determine which features and feature combinations best differentiate competitive overlaps from non-competitive overlaps. It was found that overlap placement features played a greater role than prosodic features in indicating turn competition. Among the prosodic features tested, F0 and intensity were the most effective predictors of turn competition. Also, our decision tree models suggest that turn competitive and non-competitive overlaps can be initiated by a new speaker at many different points in the current speakers turn. These findings have implications for the design of dialogue systems, and suggest novel hypotheses about how speakers deploy phonetic resources in everyday talk.

A Graph-Based Approach to Topic Clustering for Online Comments to News

This paper investigates graph-based approaches to labeled topic clustering of reader comments in online news. For graph-based clustering we propose a linear regression model of similarity between the graph nodes (comments) based on similarity features and weights trained using automatically derived training data. To label the clusters our graph-based approach makes use of DBPedia to abstract topics extracted from the clusters. We evaluate the clustering approach against gold standard data created by human annotators and compare its results against LDA – currently reported as the best method for the news comment clustering task. Evaluation of cluster labelling is set up as a retrieval task, where human annotators are asked to identify the best cluster given a cluster label. Our clustering approach significantly outperforms the LDA baseline and our evaluation of abstract cluster labels shows that graph-based approaches are a promising method of creating labeled clusters of news comments, although we still find cases where the automatically generated abstractive labels are insufficient to allow humans to correctly associate a label with its cluster.

Fundamental frequency height as a resource for the management of overlap in talk-in-interaction.

Overlapping talk is common in talk-in-interaction. Much of the previous research on this topic agrees that speaker overlaps can be either turn competitive or noncompetitive. An investigation of the differences in prosodic design between these two classes of overlaps can offer insight into how speakers use and orient to prosody as a resource for turn competition. In this paper, we investigate the role of fundamental frequency (F0) as a resource for turn competition in overlapping speech. Our methodological approach combines detailed conversation analysis of overlap instances with acoustic measurements of F0 in the overlapping sequence and in its local context. The analyses are based on a collection of overlap instances drawn from the ICSI Meeting corpus. We found that overlappers mark an overlapping incoming as competitive by raising F0 above their norm for turn beginnings, and retaining this higher F0 until the point of overlap resolution. Overlappees may respond to these competitive incomings by returning competition, in which case they raise their F0 too. Our results thus provide instrumental support for earlier claims made on impressionistic evidence, namely that participants in talk-in-interaction systematically manipulate F0 height when competing for the turn.

The SENSEI Annotated Corpus: Human Summaries of Reader Comment Conversations in On-line News

Researchers are beginning to explore how to generate summaries of extended argumentative conversations in social media, such as those found in reader comments in on-line news. To date, however, there has been little discussion of what these summaries should be like and a lack of humanauthored exemplars, quite likely because writing summaries of this kind of interchange is so difficult. In this paper we propose one type of reader comment summary – the conversation overview summary – that aims to capture the key argumentative content of a reader comment conversation. We describe a method we have developed to support humans in authoring conversation overview summaries and present a publicly available corpus – the first of its kind – of news articles plus comment sets, each multiply annotated, according to our method, with conversation overview summaries.

Comment-to-Article Linking in the Online News Domain

Online commenting to news articles provides a communication channel between media professionals and readers offering a crucial tool for opinion exchange and freedom of expression. Currently, comments are detached from the news article and thus removed from the context that they were written for. In this work, we propose a method to connect readers’ comments to the news article segments they refer to. We use similarity features to link comments to relevant article segments and evaluate both word-based and term-based vector spaces. Our results are comparable to state-of-theart topic modeling techniques when used for linking tasks. We demonstrate that article segments and comments representation are relevant to linking accuracy since we achieve better performances when similarity features are computed using similarity between terms rather than words.

Automatic Label Generation for News Comment Clusters

We present a supervised approach to automat- ically labelling topic clusters of reader com- ments to online news. We use a feature set that includes both features capturing proper- ties local to the cluster and features that cap- ture aspects from the news article and from comments outside the cluster. We evaluate the approach in an automatic and a manual, task-based setting. Both evaluations show the approach to outperform a baseline method, which uses tf*idf to select comment-internal terms for use as topic labels. We illustrate how cluster labels can be used to generate cluster summaries and present two alternative sum- mary formats: a pie chart summary and an ab- stractive summary.