Discourse Coherence

25.2 Discourse structure parsing

1. What is discourse parsing?
   Automatically determining the coherence relations between sentences: a full tree for RST, only labeled span pairs for PDTB.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, pp. 574–575
2. How is EDU segmentation done?
   Early systems post-processed syntactic parses (EDUs are roughly clauses). Modern systems are neural sequence models: an encoder plus a linear layer and softmax label each token 1 if it starts an EDU, else 0, supervised by gold segmentations.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 575
3. What are the actions of the Yu et al. (2018) shift-reduce RST parser?
   shift: push the first EDU of the queue onto the stack as a single-node subtree. reduce(l, d): merge the top two subtrees with relation label \(l\) and nuclearity \(d \in \{NN, NS, SN\}\). pop root: remove the final tree.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 575
4. Give the shift-reduce actions for an RST tree with EDUs e1 (“AT&T said it”), e2 (“will lay off technicians”), e3 (“The workers install… exchanges,”), e4 (“which are large networks.”).

   action	stack after
   SH	e1
   SH	e1, e2
   RD(attr, SN)	e1:2
   SH	e1:2, e3
   SH	e1:2, e3, e4
   RD(elab, NS)	e1:2, e3:4
   RD(elab, SN)	e1:4
   PR	(done)

   The nuclearity says which side is the nucleus: SN = satellite first (e1 is the attribution satellite of e2), NS = nucleus first (e4 elaborates e3).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 576
5. How does the Yu et al. RST parser represent its state and choose actions?
   A hierarchical biLSTM encodes words within each EDU and then the EDU sequence; EDU and subtree representations are averages of encoder outputs. A feedforward decoder scores actions from the top three stack subtrees plus the first queue EDU.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 576
6. How is the Yu et al. RST parser trained?
   Map each gold tree to a sequence of oracle actions and train with cross-entropy over action probabilities (plus L2 regularization).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 577
7. How are RST parsers evaluated?
   On the RST Discourse Treebank test set, with gold EDUs or end-to-end, by RST-Pareval metrics on binarized trees: span F1 (S), nuclearity (N), relation (R) and full (F), micro-averaged over spans.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 577
8. What is shallow discourse parsing, and what are its four subtasks?
   PDTB parsing: flat relations between spans, not trees. Subtasks (Lin et al.): (1) find discourse connectives, (2) find each connective’s two spans, (3) label the relation, (4) assign a relation to every adjacent sentence pair without an explicit connective.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 577
9. What is a simple strong method for PDTB implicit relation classification (Task 4)?
   Encode the two spans with BERT, take the [CLS] hidden state, pass it through a one-layer tanh feedforward network and a softmax over the 11 second-level senses (Nie et al. 2019).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 577
10. Why is detecting discourse connectives a disambiguation problem?
    The same words have non-discourse uses: and links clauses in an expansion relation, but is NP conjunction in “blue and green”; once is temporal in “once it enters the lungs” but means ‘formerly’ in “once used to make filters”. It is a kind of word sense disambiguation, today done with biLSTM-CRF BIO tagging.
    • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, pp. 577–578
11. How are PDTB argument spans found?
    With sequence models (e.g. BERT inputs to a biLSTM), or simple heuristics: almost all relations lie within one sentence or span two adjacent sentences with one argument in each.
    • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 25, §25.2, p. 578