Dan Roth · page 2
Papers
- Zero-shot Event Extraction via Transfer Learning: Challenges and Insights (2021 · 128 citations)
Event extraction (EE) usually depends on costly supervised annotation tied to a fixed event ontology. This paper tests whether EE can be done with no event training data at all. The authors reuse off-the-shelf pretrained models built for other tasks. Trigger detection is posed as Textual Entailment or Yes/No QA… - Grammar Error Correction in Morphologically Rich Languages: The Case of Russian (2019 · 126 citations)
Research on grammar error correction (GEC) has focused almost entirely on English, where systems rely on tens of millions of words of annotated learner text. - Solving Hard Coreference Problems (2015 · 118 citations)
Coreference systems still struggle with pronouns, especially in Winograd-style sentences. In these, gender and number agreement offer no help, and picking the right antecedent depends on world knowledge. Peng, Khashabi and Roth propose Predicate Schemas, a knowledge representation organized around predicates: - A Statistical Analysis of Summarization Evaluation Metrics Using Resampling Methods (2021 · 117 citations)
Automatic summarization metrics are usually judged by how well their scores correlate with human quality ratings. - TwoWingOS: A Two-Wing Optimization Strategy for Evidential Claim Verification (2018 · 114 citations)
The paper tackles evidential claim verification: given a claim and a large text collection, a system must find the sentences that support or refute the claim and then decide whether the claim is supported, refuted, or lacks enough information. - ExpertQA: Expert-Curated Questions and Attributed Answers (2024 · 113 citations)
LLMs are increasingly used by specialists, but it is unclear how factual their answers are and how well their citations hold up in specific professional fields, especially high-risk ones like medicine and law. - “Ask Not What Textual Entailment Can Do for You...” (2010 · 108 citations)
The paper argues that Recognizing Textual Entailment (RTE) should become a shared testbed for the component tasks of NLP, such as NER, coreference and parsing. - The Illinois-Columbia System in the CoNLL-2014 Shared Task (2014 · 94 citations)
This paper presents the Illinois-Columbia entry to the CoNLL-2014 shared task on correcting grammatical mistakes in essays by non-native English writers. - CogCompTime: A Tool for Understanding Time in Natural Language (2018 · 93 citations)
CogCompTime is a publicly available system for extracting temporal information from raw English text. It covers two tasks. The first is finding explicit time expressions (Timexes) and converting them to a standard format. The second is recovering the implicit ordering of events through temporal relations (TempRels). - A Constrained Latent Variable Model for Coreference Resolution (2013 · 81 citations)
Coreference resolution clusters noun-phrase mentions according to which real-world entity each one denotes. - Modeling Semantic Relations Expressed by Prepositions (2013 · 76 citations)
Prepositions are highly ambiguous: one preposition can signal many meanings, and many prepositions can signal the same meaning. - A Joint Framework for Coreference Resolution and Mention Head Detection (2015 · 75 citations)
Coreference systems lose much of their accuracy when they must find mentions in raw text rather than receive gold-annotated ones. - "I'd rather just go to bed": Understanding Indirect Answers (2020 · 72 citations)
The paper builds Circa, 34,268 pairs of a polar question and an indirect answer (no yes/no words), in ten everyday scenarios, written and labelled by crowd workers. - Learning-based Multi-Sieve Co-reference Resolution with Knowledge (2012 · 72 citations)
The paper tackles co-reference resolution: deciding which mentions in a text point to the same entity. Many errors come from missing world knowledge, such as not knowing that "Kursk" is a submarine and can therefore be called a "vessel." The authors link mentions to Wikipedia pages using an existing context-sensitive… - A Peek into Token Bias: Large Language Models Are Not Yet Genuine Reasoners (2024 · 64 citations)
The paper asks whether large language models actually reason logically or mostly react to familiar surface tokens that appear in well-known problem statements. - Inference Protocols for Coreference Resolution (2011 · 50 citations)
This paper presents Illinois-Coref, the system the University of Illinois submitted to the closed-track competition of CoNLL-2011, a shared task on resolving coreference in OntoNotes-4.0. - CoCoMIC: Code Completion by Jointly Modeling In-file and Cross-file Context (2024 · 45 citations)
Most code language models complete a file using only the code above the cursor. They ignore definitions elsewhere in the same repository, which matter a great deal in modular projects. Without that information, models invent class methods or call functions with wrong arguments, especially in private, still-developing… - QANom: Question-Answer driven SRL for Nominalizations (2020 · 41 citations)
QANom extends QA-SRL from verbs to nouns derived from verbs: a noun like proof or increase is questioned with its verb's question templates («What has someone proved?», «Why did something increase?») and answered with spans of the sentence. - DeAL: Decoding-time Alignment for Large Language Models (2025 · 36 citations)
The paper argues that alignment learned during training (e.g., RLHF) has two weaknesses. It locks models into a developer's fixed, supposedly universal preferences, and it can be bypassed, for example through jailbreaks. The authors propose DeAL, which moves alignment to inference time. Decoding is treated as an… - PropSegmEnt: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition (2023 · 31 citations)
Standard NLI datasets give one entailment label to a whole sentence or paragraph. Most real sentences pack several distinct units of meaning, and each can be true or unsupported independently. A single label therefore hides partial entailment, where only some of a hypothesis is backed by the premise. PropSegmEnt… - Pairwise Representation Learning for Event Coreference (2022 · 31 citations)
Event coreference resolution decides whether two event mentions describe the same real-world event. Earlier systems usually encoded each mention on its own and then classified the pair. As a result, a mention's vector never reflected its counterpart's context, which is a serious gap when the mentions come from… - Revisiting the Evaluation for Cross Document Event Coreference (2016 · 30 citations)
Cross document event coreference (CDEC) asks whether event mentions in different documents describe the same real-world event. - Identifying Semitic Roots: Machine Learning with Linguistic Constraints (2008 · 25 citations)
Semitic words are built from a root (usually three consonants) and a pattern, and the root is hard to recover from unpointed text because of weak paradigms, missing vowels and attached prefixes. - Improving Generalization in Coreference Resolution via Adversarial Training (2019 · 23 citations)
Neural coreference resolvers trained on CoNLL-2012 may lean on memorized entity names rather than truly generalizing. - Learning Hebrew Roots: Machine Learning with Linguistic Constraints (2004 · 23 citations)
The root of a Hebrew word (usually three consonants) is hard to find, because vowels are unwritten, particles attach without spaces, and weak roots change their shape.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/dan-roth/page/2/