Andreas Peldszus

Joint prediction in MST-style discourse parsing for argumentation mining

We introduce a new approach to argumentation mining that we applied to a parallel German/English corpus of short texts annotated with argumentation structure. We focus on structure prediction, which we break into a number of subtasks: relation identification, central claim identification, role classification, and function classification. Our new model jointly predicts different aspects of the structure by combining the different subtask predictions in the edge weights of an evidence graph; we then apply a standard MST decoding algorithm. This model not only outperforms two reasonable baselines and two datadriven models of global argument structure for the difficult subtask of relation identification, but also improves the results for central claim identification and function classification and it compares favorably to a complex mstparser pipeline.

Towards segment-based recognition of argumentation structure in short texts

Despite recent advances in discourse parsing and causality detection, the automatic recognition of argumentation structure of authentic texts is still a very challenging task. To approach this problem, we collected a small corpus of German microtexts in a text generation experiment, resulting in texts that are authentic but of controlled linguistic and rhetoric complexity. We show that trained annotators can determine the argumentation structure on these microtexts reliably. We experiment with different machine learning approaches for automatic argumentation structure recognition on various levels of granularity of the scheme. Given the complex nature of such a discourse understanding tasks, the first results presented here are promising, but invite for further investigation.

Rhetorical structure and argumentation structure in monologue text.

On the basis of a new corpus of short “microtexts” with parallel manual annotations, we study the mapping from discourse structure (in terms of Rhetorical Structure Theory, RST) to argumentation structure. We first perform a qualitative analysis and discuss our findings on correspondence patterns. Then we report on experiments with deriving argumentation structure from the (gold) RST trees, where we compare a tree transformation model, an aligner based on subgraph matching, and a more complex “evidence graph” model.

Towards Detecting Counter-considerations in Text

Argumentation mining obviously involves finding support relations between statements, but many interesting instances of argumentation also contain counter-considerations, which the author mentions in order to preempt possible objections by the readers. A counterconsideration in monologue text thus involves a switch of perspective toward an imaginary opponent. We present a classification approach to classifying counter-considerations and apply it to two different corpora: a selection of very short argumentative texts produced in a text generation experiment, and a set of newspaper commentaries. As expected, the latter pose more difficulties, which we investigate in a brief error anaylsis.