Computational semantics performs automatic meaning analysis of natural language. Research in computational semantics designs meaning representations and develops mechanisms for automatically assigning those representations and reasoning over them. Computational semantics is not a single monolithic task but consists of many subtasks, including word sense disambiguation, multi-word expression analysis, semantic role labeling, the construction of sentence semantic structure, coreference resolution, and the automatic induction of semantic information from data.
The development of manually constructed resources has been vastly important in driving the field forward. Examples include WordNet, PropBank, FrameNet, VerbNet, and TimeBank. These resources specify the linguistic structures to be targeted in automatic analysis, and they provide high-quality human-generated data that can be used to train machine learning systems. Supervised machine learning based on manually constructed resources is a widely used technique.
A second core strand has been the induction of lexical knowledge from text data. For example, words can be represented through the contexts in which they appear (called distributional vectors or embeddings), such that semantically similar words have similar representations. Or semantic relations between words can be inferred from patterns of words that link them. Wide-coverage semantic analysis always needs more data, both lexical knowledge and world knowledge, and automatic induction at least alleviates the problem.
Compositionality is a third core theme: the systematic construction of structural meaning representations of larger expressions from the meaning representations of their parts. The representations typically use logics of varying expressivity, which makes them well suited to performing automatic inferences with theorem provers.
Manual specification and automatic acquisition of knowledge are closely intertwined. Manually created resources are automatically extended or merged. The automatic induction of semantic information is guided and constrained by manually specified information, which is much more reliable. And for restricted domains, the construction of logical representations is learned from data.
It is at the intersection of manual specification and machine learning that some of the current larger questions of computational semantics are located. For instance, should we build general-purpose semantic representations, or is lexical knowledge simply too domain-specific, and would we be better off learning task-specific representations every time? When performing inference, is it more beneficial to have the solid ground of a human-generated ontology, or is it better to reason directly with text snippets for more fine-grained and gradual inference? Do we obtain a better and deeper semantic analysis as we use better and deeper manually specified linguistic knowledge, or is the future in powerful learning paradigms that learn to carry out an entire task from natural language input and output alone, without pre-specified linguistic knowledge?
12
Article
Computational Semantics
Katrin Erk
Article
Type Theory for Natural Language Semantics
Stergios Chatzikyriakidis and Robin Cooper
Type theory is a regime for classifying objects (including events) into categories called types. It was originally designed in order to overcome problems relating to the foundations of mathematics relating to Russell’s paradox. It has made an immense contribution to the study of logic and computer science and has also played a central role in formal semantics for natural languages since the initial work of Richard Montague building on the typed λ-calculus. More recently, type theories following in the tradition created by Per Martin-Löf have presented an important alternative to Montague’s type theory for semantic analysis. These more modern type theories yield a rich collection of types which take on a role of representing semantic content rather than simply structuring the universe in order to avoid paradoxes.
Article
Models of Human Sentence Comprehension in Computational Psycholinguistics
John Hale
Computational models of human sentence comprehension help researchers reason about how grammar might actually be used in the understanding process. Taking a cognitivist approach, this article relates computational psycholinguistics to neighboring fields (such as linguistics), surveys important precedents, and catalogs open problems.
Article
Learnability and Learning Algorithms in Phonology
Anne-Michelle Tessier
Phonological learnability deals with the formal properties of phonological languages and grammars, which are combined with algorithms that attempt to learn the language-specific aspects of those grammars. The classical learning task can be outlined as follows: Beginning at a predetermined initial state, the learner is exposed to positive evidence of legal strings and structures from the target language, and its goal is to reach a predetermined end state, where the grammar will produce or accept all and only the target language’s strings and structures. In addition, a phonological learner must also acquire a set of language-specific representations for morphemes, words and so on—and in many cases, the grammar and the representations must be acquired at the same time.
Phonological learnability research seeks to determine how the architecture of the grammar, and the workings of an associated learning algorithm, influence success in completing this learning task, i.e., in reaching the end-state grammar. One basic question is about convergence: Is the learning algorithm guaranteed to converge on an end-state grammar, or will it never stabilize? Is there a class of initial states, or a kind of learning data (evidence), which can prevent a learner from converging? Next is the question of success: Assuming the algorithm will reach an end state, will it match the target? In particular, will the learner ever acquire a grammar that deems grammatical a superset of the target language’s legal outputs? How can the learner avoid such superset end-state traps? Are learning biases advantageous or even crucial to success?
In assessing phonological learnability, the analysist also has many differences between potential learning algorithms to consider. At the core of any algorithm is its update rule, meaning its method(s) of changing the current grammar on the basis of evidence. Other key aspects of an algorithm include how it is triggered to learn, how it processes and/or stores the errors that it makes, and how it responds to noise or variability in the learning data. Ultimately, the choice of algorithm is also tied to the type of phonological grammar being learned, i.e., whether the generalizations being learned are couched within rules, features, parameters, constraints, rankings, and/or weightings.
Article
Computational Phonology
Jane Chandlee and Jeffrey Heinz
Computational phonology studies the nature of the computations necessary and sufficient for characterizing phonological knowledge. As a field it is informed by the theories of computation and phonology.
The computational nature of phonological knowledge is important because at a fundamental level it is about the psychological nature of memory as it pertains to phonological knowledge. Different types of phonological knowledge can be characterized as computational problems, and the solutions to these problems reveal their computational nature. In contrast to syntactic knowledge, there is clear evidence that phonological knowledge is computationally bounded to the so-called regular classes of sets and relations. These classes have multiple mathematical characterizations in terms of logic, automata, and algebra with significant implications for the nature of memory. In fact, there is evidence that phonological knowledge is bounded by particular subregular classes, with more restrictive logical, automata-theoretic, and algebraic characterizations, and thus by weaker models of memory.
Article
Game Theory in Pragmatics: Evolution, Rationality, and Reasoning
Michael Franke
Game theory provides formal means of representing and explaining action choices in social decision situations where the choices of one participant depend on the choices of another. Game theoretic pragmatics approaches language production and interpretation as a game in this sense. Patterns in language use are explained as optimal, rational, or at least nearly optimal or rational solutions to a communication problem. Three intimately related perspectives on game theoretic pragmatics are sketched here: (i) the evolutionary perspective explains language use as the outcome of some optimization process, (ii) the rationalistic perspective pictures language use as a form of rational decision-making, and (iii) the probabilistic reasoning perspective considers specifically speakers’ and listeners’ beliefs about each other. There are clear commonalities behind these three perspectives, and they may in practice blend into each other.
At the heart of game theoretic pragmatics lies the idea that speaker and listener behavior, when it comes to using a language with a given semantic meaning, are attuned to each other. By focusing on the evolutionary or rationalistic perspective, we can then give a functional account of general patterns in our pragmatic language use. The probabilistic reasoning perspective invites modeling actual speaker and listener behavior, for example, as it shows in quantitative aspects of experimental data.
Article
Textual Inference
Annie Zaenen
Hearers and readers make inferences on the basis of what they hear or read. These inferences are partly determined by the linguistic form that the writer or speaker chooses to give to her utterance. The inferences can be about the state of the world that the speaker or writer wants the hearer or reader to conclude are pertinent, or they can be about the attitude of the speaker or writer vis-à-vis this state of affairs. The attention here goes to the inferences of the first type. Research in semantics and pragmatics has isolated a number of linguistic phenomena that make specific contributions to the process of inference. Broadly, entailments of asserted material, presuppositions (e.g., factive constructions), and invited inferences (especially scalar implicatures) can be distinguished.
While we make these inferences all the time, they have been studied piecemeal only in theoretical linguistics. When attempts are made to build natural language understanding systems, the need for a more systematic and wholesale approach to the problem is felt. Some of the approaches developed in Natural Language Processing are based on linguistic insights, whereas others use methods that do not require (full) semantic analysis.
In this article, I give an overview of the main linguistic issues and of a variety of computational approaches, especially those stimulated by the RTE challenges first proposed in 2004.
Article
Generative Grammar
Knut Tarald Taraldsen
This article presents different types of generative grammar that can be used as models of natural languages focusing on a small subset of all the systems that have been devised. The central idea behind generative grammar may be rendered in the words of Richard Montague: “I reject the contention that an important theoretical difference exists between formal and natural languages” (“Universal Grammar,” Theoria, 36 [1970], 373–398).
Article
Computer-Based Tools for Word and Paradigm Computational Morphology
Raphael Finkel
The Word and Paradigm approach to morphology associates lexemes with tables of surface forms for different morphosyntactic property sets. Researchers express their realizational theories, which show how to derive these surface forms, using formalisms such as Network Morphology and Paradigm Function Morphology. The tables of surface forms also lend themselves to a study of the implicative theories, which infer the realizations in some cells of the inflectional system from the realizations of other cells.
There is an art to building realizational theories. First, the theories should be correct, that is, they should generate the right surface forms. Second, they should be elegant, which is much harder to capture, but includes the desiderata of simplicity and expressiveness. Without software to test a realizational theory, it is easy to sacrifice correctness for elegance. Therefore, software that takes a realizational theory and generates surface forms is an essential part of any theorist’s toolbox.
Discovering implicative rules that connect the cells in an inflectional system is often quite difficult. Some rules are immediately apparent, but others can be subtle. Software that automatically analyzes an entire table of surface forms for many lexemes can help automate the discovery process.
Researchers can use Web-based computerized tools to test their realizational theories and to discover implicative rules.
12
