About

Steven Abney is a professor in the Department of Linguistics at the University of Michigan, with an affiliation in Computer Science and Engineering and the Michigan Institute for Data Science. He earned his Ph.D. from MIT in 1987. His area of research is computational linguistics, encompassing language technology such as machine translation, speech recognition, and information extraction, as well as digital linguistics, the language component of artificial intelligence, and computational psycholinguistics. Abney's career has alternated between academic linguistic departments and computer science departments in industrial research labs. He views language as an intrinsically computational system and considers computational linguistics to be a core part of linguistics. He emphasizes that languages are as complex as subatomic particles, galaxies, or living cells, and advocates for studying them with the mathematical and computational sophistication common in physics, astronomy, or molecular biology. His current projects include language digitization, low-resource computational linguistics, digital documentation of Ojibwe, semantic interpretation, dynamic semantics, and improper anaphora. He has also worked on dependency parsing, semisupervised learning, spectral methods, information extraction (notably for biomedicine), partial and deterministic parsing, grammatical inference, conversational agents, spoken language systems, automated phonetic transcription, and automated harmonic analysis of music. Abney teaches courses such as Mathematics of Language, Computational Linguistics, Machine Learning for NLP, and undergraduate and graduate courses in semantics.

Research topics

• Computer Science
• Linguistics
• Natural Language Processing
• Programming language
• Philosophy
• Psychology
• Cognitive science
• Epistemology

Selected publications

• Intensional anaphora

  Semantics and Pragmatics · 2024-07-14 · 1 citations

  articleOpen accessSenior author

  Intensional operators are often treated as quantifiers over possible worlds, parallel to the treatment of determiners as quantifiers over individuals. Individuals introduced in intensional contexts cannot serve as antecedents to later pronouns as easily as those introduced in (merely) quantificational contexts, though. For instance, a quantified sentence like Everyone is eating a cheeseburger may be felicitously followed by an anaphoric statement like They are large, where they refers to the totality of cheeseburgers being eaten. However, as Stone (1999) points out, the quite similar Andrea might be eating a cheeseburger does not support later anaphoric references such as It is large or They are large. Stone (1999), Stone & Hardt (1999), and Brasoveanu (2010) solve this problem by restricting the value of pronouns: in their systems, a pronoun presupposes that its referent(s) exist in the world of evaluation, ruling out anaphora from non-veridical intensional contexts. And yet, we show in this paper (i) cases where such anaphora is disallowed even when the pronoun’s referents clearly exist and (ii) cases where such anaphora is indeed allowed, even though the pronoun’s referents might not exist. We argue instead that intensional anaphora is best captured using a description-based, rather than a value-based account. We propose that a pronoun presupposes that its corresponding antecedent description is instantiated in each world of the context set. For instance, there must be a cheeseburger being eaten by Andrea in every candidate world of the context set in order for It is large to be felicitous after Andrea might be eating a cheeseburger. We implement our proposal via a new logic (building on work by Keshet 2018, Abney & Keshet 2022) that we name Plural Intensional Presuppositional predicate calculus (or PIP). Each PIP formula translates directly into standard first-order predicate calculus with set abstraction, providing a classical foundation for this work. EARLY ACCESS

  PublisherOA PDFDOI

• Inductive general grammar

  Glossa a journal of general linguistics · 2021 · 3 citations

  1st authorCorresponding
  • Computer Science
  • Linguistics
  • Computer Science

  General-linguistic datasets that have become available in recent years promise to enable new progress toward a theory of general grammar. A barrier to success is the incompatibility between the inductive, externalist approach that is natural for exploiting the datasets and the deductive, mentalist philosophy that is currently dominant within linguistics. I argue that the externalist philosophy is viable and that there are reasons to consider it preferable. I argue that the mainstream approach is in some cases unnecessarily concerned with psychological reality, and in other cases too quick to reject required subtheories on the grounds that they belong to “language processing” rather than linguistics, with the result that current grammars give systematically inaccurate answers to questions of the linguistic status of sentences. I suggest that the inductive development of general grammar is already being carried out (though not in those terms) in the field of natural language processing, and that linguistic participation in the effort would be of benefit to both fields.

  PublisherOA PDFDOI

• A Summary of the First Workshop on Language Technology for Language\n Documentation and Revitalization

  arXiv (Cornell University) · 2020 · 5 citations

  • Computer Science
  • Linguistics
  • Computer Science

  Despite recent advances in natural language processing and other language\ntechnology, the application of such technology to language documentation and\nconservation has been limited. In August 2019, a workshop was held at Carnegie\nMellon University in Pittsburgh to attempt to bring together language community\nmembers, documentary linguists, and technologists to discuss how to bridge this\ngap and create prototypes of novel and practical language revitalization\ntechnologies. This paper reports the results of this workshop, including issues\ndiscussed, and various conceived and implemented technologies for nine\nlanguages: Arapaho, Cayuga, Inuktitut, Irish Gaelic, Kidaw'ida, Kwak'wala,\nOjibwe, San Juan Quiahije Chatino, and Seneca.\n

  PublisherOA PDFDOI

• A Summary of the First Workshop on Language Technology for Language Documentation and Revitalization

  Workshop Spoken Language Technologies for Under-resourced Languages · 2020 · 1 citations

  • Computer Science
  • Computer Science
  • Natural Language Processing

  Despite recent advances in natural language processing and other language technology, the application of such technology to language documentation and conservation has been limited. In August 2019, a workshop was held at Carnegie Mellon University in Pittsburgh to attempt to bring together language community members, documentary linguists, and technologists to discuss how to bridge this gap and create prototypes of novel and practical language revitalization technologies. This paper reports the results of this workshop, including issues discussed, and various conceived and implemented technologies for nine languages: Arapaho, Cayuga, Inuktitut, Irish Gaelic, Kidaw'ida, Kwak'wala, Ojibwe, San Juan Quiahije Chatino, and Seneca.

  PublisherOA PDF

• A bidirectional mapping between English and CNF-based reasoners

  ScholarWorks@UMassAmherst (University of Massachusetts Amherst) · 2018-01-01 · 1 citations

  articleOpen access1st authorCorresponding

  If language is a transduction between sound and meaning, the target of semantic interpretation should be the meaning representation expected by general cognition. Automated reasoners provide the best available fully-explicit proxies for general cognition, and they commonly expect Clause Normal Form (CNF) as input. There is a well-known algorithm for converting from unrestricted predicate calculus to CNF, but it is not invertible, leaving us without a means to transduce CNF back to English. I present a solution, with possible repercussions for the overall framework of semantic interpretation.

  PublisherOA PDFDOI

• Experiments in Sentence Language Identification with Groups of Similar Languages

  2014-01-01 · 24 citations

  articleOpen accessSenior author

  Language identification is a simple problem that becomes much more difficult when its usual assumptions are broken. In this paper we consider the task of classifying short segments of text in closely-related languages for the Discriminating Similar Languages shared task, which is broken

  PublisherDOI

• Anders Søgaard: Semi-Supervised Learning and Domain Adaptation in Natural Language Processing

  Machine Translation · 2014-01-31 · 1 citations

  article1st authorCorresponding
  PublisherDOI

• Labeling the Languages of Words in Mixed-Language Documents using Weakly Supervised Methods

  2013-01-01 · 153 citations

  articleSenior author

  In this paper we consider the problem of labeling the languages of words in mixed-language documents. This problem is approached in a weakly supervised fashion, as a sequence labeling problem with monolingual text samples for training data. Among the approaches evaluated, a conditional random field model trained with generalized expectation criteria was the most accurate and performed consistently as the amount of training data was varied. 1

  Publisher

• Morphological inference from bitext for resource-poor languages

  2012-01-01 · 4 citations

  dissertation1st authorCorresponding

  The development of rich, multi-lingual corpora is essential for enabling new types of large-scale inquiry into the nature of language (Abney and Bird, 2010; Lewis and Xia, 2010). However, significant digital resources currently exist for only a handful of the world's languages. The present dissertation addresses this issue by introducing new techniques for creating rich corpora by enriching existing resources via automated processing. As a way of leveraging existing resources, this dissertation describes an automated method for extracting bitext (text accompanied by a translation) from bilingual documents. Digitized copies of printed books are mined for foreign-language material, using statistical methods for language identification and word alignment to identify instances of English-foreign bitext. After parsing the English text and transferring this analysis via the word alignments, the foreign word tokens are tagged with English glosses and morphosyntactic features. Tagged tokens such as these constitute the input to a new algorithm, presented in this dissertation, for performing morphology induction. Drawing on previous work on unsupervised morphology induction which uses the principle of minimum description length to drive the analysis (Goldsmith, 2001), the present algorithm uses a greedy hill-climbing search to minimize the size of a paradigm-based morphological description of the language. The algorithm simultaneously segments wordforms into their component morphemes and organizes stems and affixes into a paradigmatic structure. Because tagged tokens are used as input, the morphemes produced by this induction method are paired with meaningful morphosyntactic features, an improvement over algorithms for unsupervised morphology based on monolingual text, which treat morphemes purely as strings of letters. Combined, these methods for collecting and analyzing bitext data offer a pathway for the automatic creation of richly-annotated corpora for resource-poor languages, requiring minimal amounts of data and minimal manual analysis.

  Publisher

• Data-Intensive Experimental Linguistics

  Linguistic Issues in Language Technology · 2011-10-01 · 16 citations

  articleOpen access1st authorCorresponding

  Computational linguistics is not a specialization of linguistics at all; it is a branch of computer science. A large majority of computational linguists have degrees in computer science and positions in computer science departments. It was founded as an offshoot of an engineering discipline (machine translation), and has been subsequently shaped by its place within artificial intelligence, and by a heavy influx of theory and method from speech recognition (another engineering discipline) and machine learning. But computation is a means to an end; the essential feature is data collection, analysis, and prediction on the large scale. I will call it data-intensive experimental linguistics. I wish to explain how data-intensive linguistics differs from mainstream practice, why I consider it to be genuine linguistics, and why I believe that it enables fundamental advances in our understanding of language.

  PublisherOA PDFDOI

Frequent coauthors

• Victoria Fossum

  4 shared

• Donald Hindle

  4 shared

• Marc Light

  Thomson Reuters (United States)

  3 shared

• Tomek Strzalkowski

  3 shared

• Yang Ye

  Chaohu Hospital of Anhui Medical University

  3 shared

• Robert Ingria

  3 shared

• Ted Briscoe

  2 shared

• Nicoletta Calzolari

  Institute for Computational Linguistics “A. Zampolli”

  2 shared