Christopher D. Manning · page 3
Papers
- Enhanced English Universal Dependencies: An Improved Representation for Natural Language Understanding Tasks (2016 · 337 citations)
Shallow understanding tasks such as relation extraction need direct links between content words, which a strict basic tree lacks: in "Fred started to laugh" the controlled verb has no subject edge. - Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection (2020 · 325 citations)
The paper describes the UD v2 guidelines and the state of the collection at release 2.5: 157 treebanks in 90 languages, built by more than 300 contributors. - Finding Contradictions in Text (2008 · 300 citations)
The paper addresses automatic detection of conflicting statements across texts. The authors argue this task differs from textual entailment and is harder. They replace the strict logical notion of contradiction with a practical one: two sentences contradict when they are highly unlikely to be true together, provided… - Modeling Biological Processes for Reading Comprehension (2014 · 292 citations)
Most machine reading research harvests facts from huge, redundant web collections. This paper instead targets micro-reading: answering questions about one paragraph, where the answer depends on how events and entities in a described process relate to each other. The authors build ProcessBank, a set of 200 biology… - BAM! Born-Again Multi-Task Networks for Natural Language Understanding (2019 · 285 citations)
Multi-task neural models in NLP often fail to match the accuracy of separately trained single-task models. - Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (2024 · 271 citations)
Legal technology vendors have marketed their AI research assistants, which are built on retrieval-augmented generation (RAG), as largely or fully free of fabricated content. - Simpler but More Accurate Semantic Dependency Parsing (2018 · 271 citations)
Semantic dependency schemes represent meaning-oriented relations between words as directed acyclic graphs rather than trees. - Joint Parsing and Named Entity Recognition (2009 · 264 citations)
Large NLP systems often run a named entity recognizer and a parser independently. Their outputs can then disagree: an entity span may cross constituent brackets, which makes the two annotations hard to combine. Finkel and Manning address this with a single model that does constituency parsing and NER together. They… - Efficient, Feature-based, Conditional Random Field Parsing (2008 · 261 citations)
Constituency parsing in 2008 was still dominated by generative PCFG-style models. Earlier discriminative parsers that used dynamic programming either trained only on very short sentences or used no features at all. This paper presents a CRF-based parser (CRF-CFG) that scores each rule application with arbitrary… - Entity-Centric Coreference Resolution with Model Stacking (2015 · 253 citations)
Coreference systems that score pairs of mentions are accurate and simple to train, but they ignore what is already known about whole entities. - Joint Learning Improves Semantic Role Labeling (2005 · 251 citations)
Semantic role labeling (SRL) systems have mostly labeled each parse-tree node on its own, sometimes adding a separate label-sequence model. - Optimizing the Factual Correctness of a Summary: A Study of Summarizing Radiology Reports (2020 · 241 citations)
Abstractive summarizers trained to match human references can score well on overlap metrics like ROUGE and still state wrong facts. - Event Extraction as Dependency Parsing (2011 · 236 citations)
The paper addresses biomedical event extraction, where events can take other events as arguments and so form nested structures. - Better Arabic Parsing: Baselines, Evaluations, and Analysis (2010 · 232 citations)
Green and Manning ask why constituency parsers for Modern Standard Arabic lag behind English results on the Penn Arabic Treebank (ATB). - A Global Joint Model for Semantic Role Labeling (2008 · 214 citations)
Most semantic role labeling (SRL) systems label each parse-tree constituent separately. This ignores a basic linguistic fact: a verb's arguments form a frame whose members constrain one another. For example, an agent rarely appears twice, and one argument's role can depend on how the others are realized. Toutanova… - The LinGO Redwoods Treebank: Motivation and Preliminary Applications (2002 · 214 citations)
The paper proposes a treebank that stores full HPSG analyses from a broad-coverage grammar (the LinGO English Resource Grammar) instead of one fixed tree format, so that users can extract the representation they need. - Did It Happen? The Pragmatic Complexity of Veridicality Assessment (2012 · 202 citations)
Event and relation extraction systems usually treat any relation they extract as true. They largely ignore veridicality, meaning whether a reader takes a described event to have actually happened. Earlier work mostly treated veridicality as a fixed lexical property, such as believe being non-veridical or realize… - LinGO Redwoods: A Rich and Dynamic Treebank for HPSG (2004 · 200 citations)
Not read: this card is written from what is reliably known about the paper, and every idea is marked to check against the text. - RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval (2024 · 193 citations)
Most retrieval-augmented language models fetch a few short, adjacent text fragments. That works poorly for questions that need a view of a whole long document, such as a book or a full scientific paper. RAPTOR addresses this by building a hierarchical index over the corpus. The text is split into pieces of about 100… - ContractNLI: A Dataset for Document-level Natural Language Inference for Contracts (2021 · 177 citations)
Contract review is slow and expensive, and people who cannot afford lawyers often sign agreements without expert checking. - Parsing to Stanford Dependencies: Trade-offs between Speed and Accuracy (2010 · 158 citations)
The authors compare ways of producing Stanford Dependencies quickly. One group is constituency parsers (Stanford, Charniak, Charniak-Johnson reranker, Bikel, Berkeley) followed by a conversion to dependencies; the other is dependency parsers trained directly on the basic dependencies (MaltParser's Nivre variants… - Answering Complex Open-domain Questions Through Iterative Query Generation (2019 · 155 citations)
Standard open-domain QA systems run one retrieval step using the question and then read the results. This fails for multi-hop questions, where a key entity, such as the author of a book, never appears in the question and cannot be found directly. The authors propose GoldEn (Gold Entity) Retriever. It alternates… - NaturalLI: Natural Logic Inference for Common Sense Reasoning (2014 · 150 citations)
The paper addresses database completion for common-sense knowledge: deciding whether a short, unseen fact (e.g., that cats have tails) should be accepted as true, given a huge collection of known facts in plain text. - Enforcing Transitivity in Coreference Resolution (2008 · 128 citations)
Pairwise coreference classifiers judge each pair of mentions separately. They then merge links after the fact, which can yield implausible chains. For example, "John Smith" and "Jane Smith" may end up in one entity only because each is linked to "Smith". Finkel and Manning argue that earlier ILP-based coreference… - Fine-tuning Language Models for Factuality (2023 · 127 citations)
LLMs often produce fluent but false statements, especially in long-form text. Checking such output by hand is slow and costly, so supervising factuality with human labels does not scale. The authors fine-tune Llama-2 7B models toward fewer factual errors without any human annotation. For each unlabeled prompt, such…
This page in Markdown → https://md.research.elamur.ai/free/en/authors/christopher-d-manning/page/3/