The Statistics of Word Cooccur-rences: Word Pairs and Collocations

By Stefan Evert. 2004

623 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/dda43b361b7d893e141cc3153bf9d4f64619a6d5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Leveraging Linguistic Structure For Open Domain Information Extraction (2015, background): “KBP relation pairs which have a high PMI2 value (Béatrice, 1994; Evert, 2005) conditioned on their type signatures matching.”
Normalized (pointwise) mutual information in collocation extraction (2009, methodology): “The first dataset contains 1212 adjective-noun bigrammes sampled from the Frankfurter Rundschau (Evert, 2008a).”
How we BLESSed distributional semantic evaluation (2011, neutral): “With this approach, we want to verify that the unequal frequency distribution across relations (see (Evert, 2005; Baroni and Lenci, 2010) to weight raw co-occurrence counts (this association measure is obtained by multiplying the raw count by Pointwise Mutual Information, and it is a close…”
Distributional Semantics in Technicolor (2012, methodology): “It is based on a word-by-document matrix, recording the distribution of the 1 http://wacky.sslmit.unibo.it/ 2 LMI is obtained by multiplying raw counts by Pointwise Mutual Information, and it is a close approximation to the Log-Likelihood Ratio (Evert, 2005).”
Don’t count, predict! A systematic comparison of context-counting vs. context-predicting semantic vectors (2014, methodology): “We considered two weighting schemes: positive Pointwise Mutual Information and Local Mutual Information (akin to the widely used Log-Likelihood Ratio scheme) (Evert, 2005).”
Evidence-based Verification for Real World Information Needs (2021, methodology): “Following (Schuster et al., 2019), we measure this phenomenon by computing local mutual information (LMI) (Evert, 2005) between the veracity label, l , claim bigrams, w , with more than 50 occurrences.”
Text: now in 2D! A framework for lexical expansion with contextual similarity (2013, background): “For a comparison of measures, see e.g. Evert (2005) and Bordag (2008).”
Few-shot Text Classification with Distributional Signatures (2019, methodology): “We compute the local mutual information (LMI) (Evert, 2005) between words and classes.”
Usage‐Based Language: Investigating the Latent Structures That Underpin Acquisition (2013, background): “Many language events across scales of analysis follow his power law: words (Evert, 2005), grammatical constructs (Ninio, 2006; O'Donnell & Ellis, 2010), formulaic phrases (O'Donnell & Ellis, 2009), and so forth.”
Towards Debiasing Fact Verification Models (2019, methodology): “LMI between w and l is deﬁned as follows: where p ( l | w ) is estimated by count( w,l ) count( w ) , p ( l ) by count( l ) | D | , p ( w, l ) by count( w,l )”
Is this a wampimuk? Cross-modal mapping between distributional semantics and the visual world (2014, methodology): “We apply Local Mutual Information (LMI, (Evert, 2005)) as weighting scheme and reduce the full co-occurrence space to 300 dimensions using the Singular Value Decomposition.”
Distributional Memory: A General Framework for Corpus-Based Semantics (2010, neutral): “The measure can also be interpreted as the dominant term of average MI or as a heuristic variant of pointwise MI to avoid its bias towards overestimating the signiﬁcance of low frequency events, and it is nearly identical to the Poisson–Stirling measure (Evert 2005).”
Linked Disambiguated Distributional Semantic Networks (2016, background): “The 1000 most significant contexts per term, as determined by the LMI significance measure [8], serve as a representation for the term, and term similarity is defined as the number of common contexts.”
Lexical Sophistication as a Multidimensional Phenomenon: Relations to Second Language Lexical Proficiency, Development, and Writing Quality (2018, neutral): “(squared) divided by the expected co-occurrence of two items (Evert, 2005): MI2 = log(observed2/expected).”
Unsupervised Does Not Mean Uninterpretable: The Case for Word Sense Induction and Disambiguation (2017, methodology): “Weights are computed using the Local Mutual Information (LMI) (Evert, 2005).”