Semantic Role Labeling

22.6 Semantic role labeling

1. What is semantic role labeling (SRL)?
   Automatically finding the semantic roles of each argument of each predicate in a sentence; usually supervised learning on FrameNet or PropBank.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 506
2. How does a classic feature-based SRL algorithm work?
   Parse the sentence; find all predicates; for each predicate, classify every parse-tree node into a semantic role or NONE with a 1-of-N classifier trained on labeled data.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 506
3. Into which steps can node classification for SRL be broken down?
   Pruning (heuristically discard unlikely constituents), identification (binary: argument or NONE), and classification (1-of-N role labeling of identified arguments).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 507
4. Why do SRL systems need a global consistency step?
   Arguments aren’t independent: constituents must not overlap, and a verb can’t have two Arg0s. Local classifier probabilities are combined by Viterbi-style decoding, reranking or integer linear programming.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 508
5. List the core SRL features of Gildea and Jurafsky (2000).
   Governing predicate, phrase type, headword and its POS, the parse path from constituent to predicate (e.g. NP↑S↓VP↓VBD), voice (active/passive), linear position (before/after the predicate), subcategorization of the predicate (e.g. VP → VBD NP PP), named entity type, and first and last words of the constituent.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 508
6. Why is the parse path feature useful in SRL?
   It compactly represents the grammatical function relation between constituent and predicate: NP↑S↓VP↓VBD is a typical subject path.

   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 508
7. Why is voice an important SRL feature?
   Passive clauses link semantic roles to surface positions very differently from active ones (the subject of a passive is usually Arg1).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 508
8. How can SRL be done as neural sequence labeling?
   Given the predicate, tag spans with BIO labels per role (B-ARG0, I-ARG0, B-ARG1…, O): encode the sentence plus the predicate (e.g. after [SEP]), concatenate each token’s encoding with a predicate indicator, and classify each token with a feedforward softmax (a CRF adds little).
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 509
9. How is SRL evaluated?
   Each argument label must be assigned to exactly the right word sequence or constituent; then precision, recall and F1 (identification and classification can be scored separately), e.g. on CoNLL-2005 or CoNLL-2012.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 22, §22.6, p. 510