Benjamin Van Durme

Ideas Mova took (3)

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

• kept — Give every frame a factuality value, not only neg and cond: happened / did not happen / not asserted (wanted, planned, hoped, asked, ordered) / uncertain (modal, hedge). (Neural Models of Factuality · implication signatures (51 verbs, prag/data/signatures.tsv) set neg/irrealis on «to» complements top-down; courses 7672/2199→7704/2206 with I2 (35db4e3f7))
• kept — Let the student remember only factual frames as facts and keep the rest as wishes or plans (world/src/knowledge.rs): «The fox wanted to eat the grapes» must not answer «Did the fox eat the grapes?» with yes. (Neural Models of Factuality · the student treats an irrealis mismatch as NegDisagree; reader alone 7677/2201, with the student 7704/2206 (35db4e3f7))
• kept — A realis flag on every frame: did the event happen according to the text? Mova has neg and cond; add «not realised» for the complement of want/plan/try/hope/refuse/promise, for future and modal frames, and for imperatives, so the student does not answer «Did… (Universal Decompositional Semantics on Universal… · realis flag for complements of want/plan/try… and «if» clauses kept (35db4e3f7); modals (will/would/might/may/shall/should) as irrealis dropped: −38 tests, −2…)

Papers

1. Temporal Reasoning in Natural Language Inference (2020)
   The paper turns four temporal datasets (TempEval3, TimeBank-Dense, RED and UDS-Time) into NLI pairs about two kinds of temporal reasoning: event order and event duration.
2. Incremental Neural Coreference Resolution in Constant Memory (2020)
   The paper turns a strong all-pairs coreference model (SpanBERT, Joshi et al. 2020) into an incremental one that reads a document segment by segment and keeps a fixed-size list of entities. Each new mention is scored against entity representations (not individual earlier mentions); it joins the best entity if the…
3. Neural Models of Factuality (2018) 2 ideas kept
   Event factuality asks, for each event-denoting predicate, whether the event happened according to the author, on a scale from −3 (certainly not) to +3 (certainly did).
4. Inference is Everything: Recasting Semantic Resources into a Unified Evaluation Framework (2017)
   The paper turns existing semantic annotations into textual-inference (RTE) pairs with templates. Three resources are recast: semantic proto-roles (SPR, about 155k pairs), FrameNet paraphrases (FN+, about 155k) and definite pronoun resolution (DPR, about 3.7k). Each pair is labelled entailed or not from the original…
5. Universal Decompositional Semantics on Universal Dependencies (2016) 1 idea kept
   The Decomp project adds semantic annotation on top of UD treebanks by asking ordinary speakers simple, decomposed questions with Likert-scale answers instead of asking for labels like Agent.
6. PPDB 2.0: Better paraphrase ranking, fine-grained entailment relations, word embeddings, and style classification (2015)
   PPDB 1.0 ranked its pivot-extracted paraphrases with a fixed heuristic, and this paper replaces the ranking with a discriminative model.
7. PPDB: The Paraphrase Database (2013)
   The paper releases PPDB 1.0, a large collection of ranked paraphrases extracted from bilingual parallel corpora by pivoting (the method of Bannard and Callison-Burch).

Co-authors (17)

• Aaron Steven White ×4
• Chris Callison-Burch ×2
• Juri Ganitkevitch ×2
• Pushpendre Rastogi ×2
• Rachel Rudinger ×2
• Adam Poliak
• Drew Reisinger
• Ellie Pavlick
• João Sedoc
• Keisuke Sakaguchi
• Kevin Duh
• Kyle Rawlins
• Patrick Xia
• Sheng Zhang
• Siddharth Vashishtha
• Tim Vieira
• Yash Kumar Lal

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/authors/benjamin-van-durme/