A Discourse Signal Annotation System for RST Trees

What it is about

The paper extends rstWeb, a browser tool for building Rhetorical Structure Theory trees, so that each relation in the tree can carry the signals that let a reader recognise it. A signal has four parts: the relation it supports, a possibly empty list of tokens (genre or layout signals have none), a type (discourse marker, semantic, syntactic, graphical, genre…) and a subtype (e.g. lexical chain, repetition). Earlier signal corpora either had no tree (PDTB) or did not anchor signals to tokens (the RST Signalling Corpus). On three re-annotated RST-SC documents (506 tokens, ~90 signals) the anchoring agreement rose from 86.0% / kappa 0.52 with a grid tool to 90.9% / 0.77 with the new interface. The tool was then used to add signals to the GUM corpus across eight genres.

Ideas

• I1. Record the signal of every link in Mova's frames, in the paper's four-part form: a cause or next link keeps the token(s) that produced it («because», «so», «and then», a purpose to) or a tokenless type (sentence order, a repeated entity). Today Frame::how explains roles but cause/next are bare indices in prag/src/frames.rs. world frames would print «cause→#1 [DM because]», and errors can be grouped by signal.
• I2. With signals recorded, report cause and next precision per signal type on the frame bench (world frames bench). Links from sentence order alone are expected to be the weakest; this tells whether to keep, restrict or drop them, with a number per type.
• I3. Tokenless signals matter in tales too: a new paragraph, a quote ending, «Once upon a time». Add them as signal types of next and check how often the bench gold has a next link across a paragraph break.

For Mova

36. Discourse relations and coherence §6 (connectives), §10 (segmentation and parsing), §11.1 (links between frames). Code: prag/src/frames.rs (link signal next to cause/next), prag/src/fbench.rs (per-signal scores). Measures: frame bench cause/next precision and recall by signal type. GUM's signal layer is NC: tests only.

Topics: Discourse analysis

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/w19-2708

@inproceedings{gessler2019discourse,
  title = {{A Discourse Signal Annotation System for RST Trees}},
  author = {Gessler, Luke and Liu, Yang and Zeldes, Amir},
  booktitle = {Proceedings of the Workshop on Discourse Relation Parsing and Treebanking 2019},
  year = {2019},
  doi = {10.18653/v1/w19-2708},
  url = {https://doi.org/10.18653/v1/w19-2708}
}

References

2 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Table 1: Comparison with Liu and Zeldes (2019).* Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. 2003. Building a discourse-tagged corpus in the framework of Rhetorical Structure Theory. In Current and New Directions in Discourse and Dialogue, Text, Speech and Language Technology 22, pages 85–112. Kluwer, Dordrecht. find our card
3. Debopam Das and Maite Taboada. 2017. Signalling of coherence relations in discourse, beyond discourse markers. Discourse Processes, pages 1–29. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/gessler-2019-discourse-signal-annotation-system-rst/