Amir Zeldes

Ideas Mova took (2)

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

• kept — Relax number agreement in a controlled way in the pronoun sieve (coref/src/sieve.rs) and in the frame reader's resolver: they/them/their may link to a singular antecedent when that antecedent is a collective noun (government, family, team, crowd, army… (A Predictive Model for Notional Anaphora in English · coref: they → singular collective noun (24 nouns) only when no strict plural fits; GUM dev gold 75.21/64.32→75.25/64.36, «they» precision 64.9→66.3, SLM trees…)
• kept — Use the paper's processing finding as an ordering rule: prefer a strictly agreeing candidate nearby, and accept a notional one only at a larger distance or when no agreeing one exists; check on GUM dev that the rule does not take links from correct plural… (A Predictive Model for Notional Anaphora in English · the ordering rule: strict agreement first, notional only as a fallback (a7fa2d765; test with a strict-plural control))

Papers

1. GUMBridge: A Corpus for Varieties of Bridging Anaphora (2026)
   GUMBridge is a new English corpus of referential bridging anaphora. It is built on the GUM corpus and covers 24 genres, written and spoken. Bridging is defined with information status: an entity is accessible on first mention because of a relation to an earlier, non-identical entity. Each bridging instance gets one…
2. What's in a Bridge?: A Descriptive, Multi-Genre Analysis of the GUMBridge Corpus for Varieties of Bridging Anaphora (2026)
   The paper is a descriptive analysis of the GUMBridge corpus: 5.3k referential bridging instances in 273k tokens across 16 genres (the train, dev and test partitions).
3. eRST: A Signaled Graph Theory of Discourse Relations and Organization (2025)
   Enhanced RST keeps the RST tree (projective, nucleus/satellite, one label per edge, over elementary discourse units) and adds two things.
4. GUM-SAGE: A Novel Dataset and Approach for Graded Entity Salience Prediction (2025)
   Salience is defined operationally: an entity is as salient as the number of one-sentence summaries of the document (out of five) that mention it, giving a 0–5 score for every named and unnamed entity of GUM's 213 documents in 12 genres.
5. Building UD Cairo for Old English in the Classroom (2025)
   The paper builds an Old English treebank of the 20 UD Cairo sentences, which illustrate the constructions of the world's languages, as a classroom exercise in historical linguistics.
6. Subjectivity in the Annotation of Bridging Anaphora (2025)
   The paper studies where human judgment enters bridging annotation, and why agreement is low. It is a pilot on the test set of the GUM corpus: 26 documents, about 26k tokens, in 16 genres, each annotated by two people. The authors separate three decisions: recognising the bridging anaphor, choosing its antecedent, and…
7. Unpacking Ambiguity: The Interaction of Polysemous Discourse Markers and Non-DM Signals (2025)
   Using GUM's eRST layer (16 genres, over 250K tokens, discourse markers plus seven types of non-marker signals: graphical, lexical, morphological, numerical, reference, semantic, syntactic), the paper measures how ambiguous each discourse marker is — the entropy of the relations it signals — and asks whether ambiguous…
8. GDTB: Genre Diverse Data for English Shallow Discourse Parsing across Modalities, Text Types, and Domains (2024)
   PDTB, the only large English shallow-discourse corpus, is news from 1989 and not openly licensed. GDTB converts the GUM corpus (16 genres, 235 documents, 228K tokens) into PDTB v3 style, using GUM's eRST relations, their connective signals, gold UD syntax and coreference: explicit relations come from eRST relations…
9. Unifying the Scope of Bridging Anaphora Types in English: Bridging Annotations in ARRAU and GUM (2024)
   The two largest English bridging resources define bridging differently. GUM/GENTLE (16 + 8 genres, 1.9k bridging links) treat it as information status: a first mention that is interpretable only through a non-identical earlier entity, with one antecedent. ARRAU (news part 229k tokens, 3.7k links) treats it as a set…
10. GENTLE: A Genre-Diverse Multilayer Challenge Set for English NLP and Linguistic Evaluation (2023)
    GENTLE is a small English challenge corpus of about 17k tokens in eight unusual genres: dictionary entries, esports commentary, legal documents, medical notes, poetry, mathematical proofs, syllabuses and threat letters.
11. The DISRPT 2023 Shared Task on Elementary Discourse Unit Segmentation, Connective Detection, and Relation Classification (2023)
    The third DISRPT task extends the 2019 and 2021 editions. It covers 26 corpora in four frameworks (RST, SDRT, PDTB and a new dependency framework, DEP) and 13 languages, of which 10 corpora are new. Three tasks are set: EDU segmentation (a BIO-style boundary label per token), connective detection (B and I labels per…
12. Are UD Treebanks Getting More Consistent? A Report Card for English UD (2023)
    The paper checks whether the two largest English UD treebanks, EWT and GUM, which have different origins and conversion histories, are converging, and whether it makes sense to train joint models on them.
13. A Second Wave of UD Hebrew Treebanking and Cross-Domain Parsing (2022)
    The paper adds a second Hebrew UD treebank, IAHLT WIKI: 5039 sentences and about 141K tokens from 39 Hebrew Wikipedia documents in seven topics (biography, events, finance, health, law, places, miscellaneous).
14. Opinion Piece: Can we Fix the Scope for Coreference? Problems and Solutions for Benchmarks beyond OntoNotes (2022)
    An argument, with examples mostly from GUM, that the OntoNotes coreference guidelines on which most systems are trained leave out cases users would expect, because they were chosen for high agreement.
15. CorefUD 1.0: Coreference Meets Universal Dependencies (2022)
    Syntax has a shared multilingual annotation standard in Universal Dependencies, but coreference resolution has had no comparable standard: corpora define the phenomena differently and use different formats, so most systems are evaluated only on English OntoNotes.
16. Mischievous nominal constructions in Universal Dependencies (2021)
    The paper looks at noun phrases that do not fit the usual UD picture: descriptors before names (titles such as Mr., role labels like "French actor"), company names with suffixes, numbered designations, dates, times and similar constructions.
17. Anatomy of OntoGUM—Adapting GUM to the OntoNotes Scheme to Evaluate Robustness of SOTA Coreference Algorithms (2021)
    GUM annotates more coreference than OntoNotes, so the authors convert it automatically to the OntoNotes scheme, using GUM's gold UD trees, in a fixed order of steps: remove bridging and cataphora (an expletive or cataphoric pronoun becomes a singleton), contract verbal mentions to the verb, merge appositions into one…
18. AMALGUM – A Free, Balanced, Multilayer English Web Corpus (2020)
    AMALGUM is an English web corpus of about 4 million tokens, built from openly licensed online sources in eight genres (news, interviews, travel guides, how-to guides, academic papers, fiction, biographies and forum posts), with about 500k tokens per genre.
19. The DISRPT 2019 Shared Task on Elementary Discourse Unit Segmentation and Connective Detection (2019)
    Discourse parsing needs a first step: cutting text into elementary discourse units (EDUs), usually clauses, and finding explicit connectives such as "because" or "in order to".
20. GumDrop at the DISRPT2019 Shared Task: A Model Stacking Approach to Discourse Unit Segmentation and Connective Detection (2019)
    GumDrop is a shared-task system for sentence splitting, elementary discourse unit segmentation and connective detection across 15 corpora in 10 languages (RST, SDRT and PDTB styles).
21. A Discourse Signal Annotation System for RST Trees (2019)
    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.
22. A Predictive Model for Notional Anaphora in English (2018) 2 ideas kept
    Notional anaphors are plural pronouns whose antecedent is grammatically singular: «the government… they», «CNN… they're on top of everything», «a number of… they», «the person here… they».
23. All Roads Lead to UD: Converting Stanford and Penn Parses to English Universal Dependencies with Multilayer Annotations (2018)
    The paper examines how to move an existing English treebank from older annotation schemes to UD 2.2, using the multilayer GUM corpus (eight genres, with entity, coreference, information-status and RST discourse layers).
24. A Characterwise Windowed Approach to Hebrew Morphological Segmentation (2018)
    Zeldes treats Hebrew segmentation as pure splitting: for every character of a space-delimited token he decides whether a boundary follows, with no morphological analysis.
25. Different Flavors of GUM: Evaluating Genre and Sentence Type Effects on Multilayer Corpus Annotation Quality (2016)
    Tagging and parsing accuracy is usually reported on newswire, and genre is the usual explanation when it falls elsewhere.

Co-authors (37)

• Lauren Levine ×6
• Siyao Peng ×5
• Yilun Zhu ×5
• Luke Gessler ×4
• Yang Janet Liu ×4
• Jessica Lin ×3
• Shabnam Behzad ×3
• Tatsuya Aoyama ×3
• Yang Liu ×3
• Debopam Das ×2
• Nathan Schneider ×2
• Anna Nedoluzhko
• Attapol Rutherford
• Chloé Braud
• Dan Simonson
• Daniel Zeman
• Devika Tiwari
• Eleni Metheniti
• Erick Galani Maziero
• Jingni Wu
• Juliano Antonio
• Junghyun Min
• Laura Rivière
• Lauren Elizabeth Levine
• Mackenzie Gong
• Martin Popel
• Michal Novák
• Mikel Iruskieta
• Nick Howell
• Noam Ordan
• Philippe Muller
• Sameer Pradhan
• Wesley Scivetti
• Yan Liu
• Yifat Ben Moshe
• Yue Yu
• Zdeněk Žabokrtský

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