Dan Klein · page 2
Papers
- Unsupervised Coreference Resolution in a Nonparametric Bayesian Model (2007 · 182 citations)
Coreference resolution is usually handled by supervised classifiers that decide whether pairs of mentions co-refer. - Better Word Alignments with Supervised ITG Models (2009 · 163 citations)
The paper addresses supervised word alignment for machine translation by combining discriminative learning with inversion transduction grammar (ITG) constraints. - Detoxifying Language Models Risks Marginalizing Minority Voices (2021 · 137 citations)
Detoxification methods aim to stop language models from producing offensive text while keeping them useful on benign input. - Parser Showdown at the Wall Street Corral: An Empirical Investigation of Error Types in Parser Output (2012 · 127 citations)
A single F-score on WSJ section 23 says nothing about which errors a constituency parser makes. The authors repair each wrong parse into the gold tree with a few tree transformations (subtree movement, node creation, node deletion), and name each repair by its linguistic type: PP attachment, clause attachment… - Ghostbuster: Detecting Text Ghostwritten by Large Language Models (2024 · 122 citations)
Ghostbuster tackles the detection of documents written by large language models when the generating model is unknown or only reachable as a black box. - Unified Pragmatic Models for Generating and Following Instructions (2018 · 115 citations)
Explicit pragmatic reasoning means agents model each other's intentions. This paper asks whether it helps in grounded instruction tasks with long, sequential actions. Earlier computational pragmatics mostly tested such reasoning on simple reference games. Fried, Andreas and Klein train two neural sequence-to-sequence… - Painless Unsupervised Learning with Features (2010 · 110 citations)
The paper argues that domain knowledge in unsupervised NLP models usually has to be encoded through conditional independence structure. - Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination (2024 · 95 citations)
This study asks whether ChatGPT treats speakers of different English dialects unequally. The authors gathered informal, native-speaker-written text in ten varieties: two prestige "standard" ones (SAE, SBE) and eight minoritized ones (AAE, Indian, Irish, Jamaican, Kenyan, Nigerian, Scottish, Singaporean English). They… - Jointly Learning to Extract and Compress (2011 · 95 citations)
The paper tackles multi-document summarization. Earlier learned systems usually trained intermediate scorers, such as per-sentence or per-word predictors, and then assembled summaries with heuristics. The authors instead train one model that directly targets final summary quality and that both selects and shortens… - DOC: Improving Long Story Coherence With Detailed Outline Control (2023 · 83 citations)
Automatically generated stories that run to several thousand words often read smoothly at the passage level but lose the overall plot, drifting away from the intended plan. - Error-Driven Analysis of Challenges in Coreference Resolution (2013 · 83 citations)
The paper gives an automatic classifier of coreference errors that works only on system output, so it applies to link-based and global-inference systems alike. - Pragmatically Informative Text Generation (2019 · 78 citations)
Conditional generators often under-inform: a restaurant description may leave out attributes of its input meaning representation. - Coreference Semantics from Web Features (2012 · 61 citations)
Bansal and Klein tackle a weakness of coreference resolution: many hard decisions depend on world knowledge, such as knowing that "Obama" can be "the president", rather than on syntax. - Decentralized Entity-Level Modeling for Coreference Resolution (2013 · 48 citations)
Coreference systems gain from entity-level information such as gender or semantic type, but exact inference over all possible partitions of mentions is intractable. - Cross-Domain Generalization of Neural Constituency Parsers (2019 · 46 citations)
The paper trains constituency parsers on the WSJ and tests them zero-shot on Brown, Genia and the English Web Treebank, and on Chinese treebank domains. - Goal Driven Discovery of Distributional Differences via Language Descriptions (2023 · 42 citations)
Comparing two large text collections by hand is slow and ad hoc. The authors frame this as an ML task called D5. A system receives a user's exploration goal and a pair of corpora. It must return natural-language predicates that describe how the corpora differ in ways that matter for that goal. For example, reports… - Tetra-Tagging: Word-Synchronous Parsing with Linear-Time Inference (2020 · 36 citations)
The paper parses constituency trees by labeling each word with one of four "tetra-tags" that encode two binary structural decisions and two labels.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/dan-klein/page/2/