A Computational Model of Non-Cooperation in
Natural Language Dialogue

Supplementary Materials

This site provides access to the materials involved in the corpus study of non-cooperation in political interviews described in Chapter 4 of the dissertation.

A list with individual links and a brief description of each resource is given below. Alternatively, these can be dowloaded as unique ZIP archive here:

Description of Materials and Transcripts

This is an extended version of Appendix A in the thesis. It includes the transcripts of the interview fragments in the entire corpus, as well as the partial annotations used as input to the second stage of annotations. Also, it reproduces the annotation guidelines for both stages in the study and the annotation tool user guide.

Corpus of Political Interviews

This archive contains the annotated corpus of political interview fragments used in the empirical study reported in Chapter 4 of the thesis. The fragments are given in plain text and in the xml format used by the annotation tool, both with and without the annotations resulting from each stage. The contents of the subfolders are described in the README file included in the archive.

Annotation Guidelines

These documents present the guidelines given to the annotators in each stage of the study. They contain relevant definitions, examples and instructions on how to carry out the annotations.

First Stage       Second Stage

Annotation Tool

The annotation was carried out using a special-purpose tool, deployed to each annotator containing the annotation data. The tool was built based on the CODA D2MTool developed at The Open University by Svetlana Stoyanchev for the CODA Project.

Among other features, the tool guides the annotators through the dataset in a fixed order and can be configured to operate according to each annotation stage. All funcionalities are described in the user guide (also part of the ZIP file above).

Algorithms for Computing Cooperation in Annotated Dialogues

This document describes the algorithms for automatically computing conversational cooperation of a dialogue, given an annotated transcript and a dialogue game.

Linguistic cooperation of a dialogue participant with respect to a conversational setting equates to the participant following the rules of the dialogue game for that conversational setting. From this perspective, each turn in a dialogue is associated with an amount of cooperation and an amount of non-cooperation, given by the number of dialogue rules, respectively, conformed with and violated in the turn. The instances in which rules are conformed with are called cooperative features and those in which rules are broken are called non-cooperative features. The degree of cooperation of each dialogue participant is thus the ratio between the number of cooperative features and the total number of features of that participant.

Contact the author for an implementation of these algorithms in Python.

Online Survey for Eliciting Human Judgement of Cooperation in Dialogue

Part of the evaluation of the measure of cooperation in dialogue was carried out by comparing the results of the method with human judgement on the interviews in the corpus (see above). This document presents a facsimile of the SurveyMonkey online survey we used to collect these judgements, and a description of how it was disseminated to attract participants.