Mark Steedman

Ideas Mova took (2)

From the ideas ledger: an idea tried, kept or queued, or measured with no gain, and what it gave.

• kept — Mova construction corpus: about 700 sentences of our own across seven constructions (object and subject relatives, reduced relatives, wh-questions, tough adjectives, free relatives), held out from training; report recall per construction for en and gparse on… (Unbounded Dependency Recovery for Parser Evaluation · en/data/constructions.tsv: 110 own sentences, 11 constructions, 220 arcs; EWT+ESLSpok model 61.8 recall, en.bin 0.4 70.0, after the parser changes 65.5…)
• kept — en parse-eval as a construction gate: a release fails if any construction falls below its last-release value even when LAS rises; check that the gate goes red on a deliberately broken tagger (the negative control). (Unbounded Dependency Recovery for Parser Evaluation · en construct-eval gate with per-model floors; negative control (tags shifted one word) turns all 11 constructions red (61.8 -> 3.6); commit 25d072835)

Papers

1. A Language-agnostic Model of Child Language Acquisition (2024)
   The authors port Abend et al.'s semantic-bootstrapping learner from English to Hebrew. The learner sees one child-directed utterance at a time, paired with a logical form, and learns a CCG lexicon with EM over Dirichlet-process counts. The Hebrew data are from the CHILDES Hagar corpus (3,295 utterances), converted…
2. Human Temporal Inferences Go Beyond Aspectual Class (2024)
   The imperfective paradox: "I was walking" entails "I walked", but "I was building a house" does not entail "I built a house".
3. Universal Dependencies and Semantics for English and Hebrew Child-directed Speech (2022)
   The authors annotate child-directed speech in English (the Adam corpus of CHILDES, 18,113 utterances) and Hebrew (the Hagar corpus, 24,172 utterances) with UD, and then convert the UD trees into logical forms, a meaning representation for acquisition models.
4. Cross-linguistically Consistent Semantic and Syntactic Annotation of Child-directed Speech (2021)
   The paper proposes a two-step method for annotating child-directed speech (CDS) in CHILDES. First, the utterances get Universal Dependencies (UD) trees, starting from the existing dependency annotation and then hand-corrected. Second, sentential logical forms (LFs) are produced from the UD trees by unlexicalised…
5. A CCG Parsing with a Supertag-factored Model (2014)
   The question is whether CCG parsing needs a full model of derivations and dependencies, or whether a model of lexical categories is enough.
6. Unbounded Dependency Recovery for Parser Evaluation (2009) 2 ideas kept
   The authors build a parser-evaluation corpus of around 700 sentences annotated with unbounded dependencies in seven grammatical constructions.
7. CCGbank: A Corpus of CCG Derivations and Dependency Structures Extracted from the Penn Treebank (2007)
   The paper gives an algorithm that translates Penn Treebank phrase-structure trees into Combinatory Categorial Grammar derivations, with binarization, head and complement/adjunct decisions, and word-word dependencies, including long-range ones from extraction, control, raising, right-node raising and coordination.
8. A Framework for Annotating Information Structure in Discourse (2005)
   The paper proposes a framework for analysing information structure in conversational English, using the Switchboard corpus.

Co-authors (18)

• Omri Abend ×3
• Ida Szubert ×2
• Louis Mahon ×2
• Nathan Schneider ×2
• Samuel Gibbon ×2
• Sharon Goldwater ×2
• Adam Lopez
• Jason Brenier
• Julia Hockenmaier
• Katarzyna Pruś
• Katherine Demuth
• Laura Rimell
• Malvina Nissim
• Mark Johnson
• Mike Lewis
• Sasha Calhoun
• Stephen Clark
• Uri Berger

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/authors/mark-steedman/