Christopher D. Manning · page 2
Papers
- Graph Convolution over Pruned Dependency Trees Improves Relation Extraction (2018 · 1,050 citations)
The paper addresses relation extraction: deciding which relation, if any, links a given subject and object entity in a sentence. - Better Word Representations with Recursive Neural Networks for Morphology (2013 · 1,027 citations)
Standard word embeddings give every surface form its own vector and ignore how words are built. As a result, rare derived words such as "distinctiveness" get poor vectors, and unseen words collapse into one or a few generic vectors. Luong, Socher and Manning instead treat morphemes (prefixes, stems, suffixes) as the… - Grounded Compositional Semantics for Finding and Describing Images with Sentences (2014 · 932 citations)
The paper addresses how to build sentence representations that reflect the visual scene a sentence describes, so that sentences and images can be retrieved from each other. - Leveraging Linguistic Structure For Open Domain Information Extraction (2015 · 923 citations)
Open IE systems turn sentences into (subject; relation; object) triples, but the usual approach of matching many learned patterns breaks on unfamiliar text and long-range dependencies. - Universal Dependencies (2021 · 910 citations)
This article sets out the linguistic theory behind Universal Dependencies (UD), which is at once a cross-linguistically consistent morphosyntactic annotation scheme, an open research community and a growing collection of treebanks. - A Thorough Examination of the CNN/Daily Mail Reading Comprehension Task (2016 · 699 citations)
The paper asks how much real language understanding the CNN/Daily Mail cloze-style reading comprehension datasets of Hermann et al. - Bilingual Word Embeddings for Phrase-Based Machine Translation (2013 · 689 citations)
Statistical MT systems struggle to recognise that two phrases mean the same thing when their exact word combinations rarely co-occur in parallel data. - Universal Stanford dependencies: A cross-linguistic typology (2014 · 627 citations)
The paper reconstructs the typology behind Stanford Dependencies and proposes a two-layer taxonomy: 42 universal grammatical relations that are broadly attested across languages, with language-specific subtypes added under them. - CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies (2017 · 604 citations)
The shared task asks systems to parse raw text, with no gold tokenization, sentence splitting or tags, into UD trees for 49 languages. - Key-Value Retrieval Networks for Task-Oriented Dialogue (2017 · 568 citations)
The paper addresses a problem of task-oriented dialogue agents. Such agents find it hard to make natural use of information from a knowledge base (KB) while remaining fully differentiable. Researchers from Stanford and Ford Research propose the Key-Value Retrieval Network. It is an LSTM-based seq2seq model with… - Part-of-Speech Tagging from 97% to 100%: Is It Time for Some Linguistics? (2011 · 511 citations)
Taggers report about 97% token accuracy, but only about 56% of sentences are fully right, and a single error can break a downstream parser. - Semi-Supervised Sequence Modeling with Cross-View Training (2018 · 494 citations)
Pre-trained representations such as word2vec or ELMo are learned from unlabeled text without any knowledge of the target task. The labeled data only enters later, during supervised training. Clark et al. propose Cross-View Training (CVT), a semi-supervised method that mixes both kinds of data throughout training. - DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature (2023 · 481 citations)
The paper asks whether a given passage came from one specific large language model. It answers without training a dedicated classifier, without collecting labelled human and machine corpora, and without watermarks. The core observation is about where a model's own samples sit: near local peaks of its log-probability… - Universal Dependencies (2021 · 480 citations)
The paper sets out the linguistic theory behind Universal Dependencies (UD, version 2). UD is often seen as a pragmatic, patchwork annotation scheme, and the authors want to show that it rests on a coherent, typologically informed view of morphology and syntax. UD treats words as the basic units, a commitment to… - Achieving Open Vocabulary Neural Machine Translation with Hybrid Word-Character Models (2016 · 475 citations)
Word-level neural MT systems usually keep a closed list of frequent words. Everything else becomes a single <unk> symbol, which is later patched by dictionary lookup or copying along attention alignments. Luong and Manning argue that this ignores two things: morphology, since rare forms are related to common ones… - A Simple and Effective Hierarchical Phrase Reordering Model (2008 · 472 citations)
Phrase-based statistical MT systems handle word-order changes poorly. Lexicalized reordering models, such as the one in Moses, only classify a phrase as monotone (M), swap (S) or discontinuous (D) relative to its immediate neighbour. Long-distance movements therefore end up in the vague D class. Galley and Manning… - Deep Reinforcement Learning for Mention-Ranking Coreference Models (2016 · 465 citations)
Coreference systems usually make many local linking decisions, but standard evaluation metrics only judge the final clustering. - Bilingual Word Representations with Monolingual Quality in Mind (2015 · 456 citations)
Earlier work on bilingual word embeddings mostly optimized for cross-lingual tasks, especially cross-lingual document classification (CLDC). - A Multi-Pass Sieve for Coreference Resolution (2010 · 448 citations)
The paper proposes an unsupervised, deterministic coreference system built as a sequence of "sieves": independent rule-based passes applied from the most precise to the least precise. - A Conditional Random Field Word Segmenter for Sighan Bakeoff 2005 (2005 · 431 citations)
The paper describes the Stanford/Colorado entry to the closed track of the SIGHAN Bakeoff 2005 Chinese word segmentation task. - Improving Coreference Resolution by Learning Entity-Level Distributed Representations (2016 · 422 citations)
Coreference resolution means grouping the mentions in a text that refer to the same entity, so it is really a clustering task. - Optimizing Chinese Word Segmentation for Machine Translation Performance (2008 · 421 citations)
The paper asks how Chinese word segmentation should be designed when its output feeds a phrase-based Chinese-to-English MT system rather than a segmentation benchmark. - Nested Named Entity Recognition (2009 · 404 citations)
Standard NER systems assume entities never overlap, so they miss entities embedded inside others, such as a location within an organization name or a protein inside a DNA mention. - A Generative Constituent-Context Model for Improved Grammar Induction (2002 · 348 citations)
The paper addresses unsupervised grammar induction: recovering unlabeled syntactic trees from part-of-speech sequences without any annotated parses. - Stanford's Graph-based Neural Dependency Parser at the CoNLL 2017 Shared Task (2017 · 346 citations)
The Stanford system is a biaffine graph parser: a BiLSTM encoder, separate ReLU layers for the head and dependent roles of arcs and labels, and biaffine scorers for arcs and labels.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/christopher-d-manning/page/2/