From a leading authority in artificial intelligence, this book delivers a synthesis of the major modern techniques and the most current research in natural language processing. The approach is unique in its coverage of semantic interpretation and discourse alongside the foundational material in syntactic processing.

This article contains the results of a theoretical analysis of the phenomenon of natural language understanding (NLU), as a methodological problem. The combination of structural-ontological and informational-psychological approaches provided an opportunity to describe the subject matter field of NLU, as a composite function of the mind, which systemically combines the verbal and discursive structural layers. In particular, the idea of NLU is presented, on the one hand, as the relation between the discourse of a specific speech message (...) and the meta-discourse of a language, in turn, activated by the need-motivational factors. On the other hand, it is conceptualized as a process with a specific structure of information metabolism, the study of which implies the necessity to differentiate the affective (emotional) and need-motivational influences on the NLU, as well as to take into account their interaction. At the same time, the hypothesis about the influence of needs on NLU under the scenario similar to the pattern of Yerkes-Dodson is argued. And the theoretical conclusion that emotions fulfill the function of the operator of the structural features of the information metabolism of NLU is substantiated. Thus, depending on the modality of emotions in the process of NLU, it was proposed to distinguish two scenarios for the implementation of information metabolism - reduction and synthetic. The argument in favor of the conclusion about the productive and constitutive role of emotions in the process of NLU is also given. (shrink)

The article argues that cognitive linguistic theory may prove an alternative to the Montague paradigm for designing natural language understanding systems. Within this framework it describes a system which models language understanding as a dialogical process between user and computer. The system operates with natural language texts as input and represent language meaning as entity-relationship diagrams.

An Example for Natural Language Understanding and the AI Problems it Raises I think this 1976 memorandum is of 1996 interest. The problems it raises haven't been solved or even substantially reformulated.

This essay considers what it means to understand natural language and whether a computer running an artificial-intelligence program designed to understand natural language does in fact do so. It is argued that a certain kind of semantics is needed to understand natural language, that this kind of semantics is mere symbol manipulation (i.e., syntax), and that, hence, it is available to AI systems. Recent arguments by Searle and Dretske to the effect that computers cannot (...) understand natural language are discussed, and a prototype natural-language-understanding system is presented as an illustration. (shrink)

��We describe a comprehensive framework for text un- derstanding, based on the representation of context. It is designed to serve as a representation of semantics for the full range of in- terpretive and inferential needs of general natural language pro- cessing. Its most distinctive feature is its uniform representation of the various simple and independent linguistic sources that play a role in determining meaning: lexical associations, syntactic re- strictions, case-role expectations, and most importantly, contextual effects. Compositional syntactic structure (...) from a shallow parsing is represented in a neural net-based associative memory, where it then interacts through a Bayesian network with semantic associa- tions and the context or “gist” of the passage carried forward from preceding sentences. Experiments with more than 2000 sentences in different languages are included. (shrink)

We describe a comprehensive framework for text un- derstanding, based on the representation of context. It is designed..

The notion of modularity, introduced by Noam Chomsky and developed with special emphasis on perceptual and linguistic processes by Jerry Fodor in his important book The Modularity of Mind, has provided a significant stimulus to research in cognitive science. This book presents essays in which a diverse group of philosophers, linguists, psycholinguists, and neuroscientists - including both proponents and critics of the modularity hypothesis - address general questions and specific problems related to modularity. Jay L. Garfield is Associate Professor of (...) Philosophy in the School of Communications and Cognitive Science at Hampshire College. (shrink)

Stephen Laurence and Eric Margolis have recently argued that certain kinds of regress arguments against the language of thought (LOT) hypothesis as an account of how we understand natural languages have been answered incorrectly or inadequately by supporters of LOT ('Regress arguments against the language of thought', Analysis, 57 (1), 60-6, J 97). They argue further that this does not undermine the LOT hypothesis, since the main sources of support for LOT are (or might be) independent of (...) it providing an account of how we understand natural language. In my paper I seek to refute both these claims, and reinstate the putative explanation of natural language understanding as a necessarily central part of the support for LOT. The main argument exploits the fact that Laurence and Margolis give too little weight to the ideas (a) that LOT might be innate (b) that for LOT supporters a semantic theory must apply to in-the-head tokens, not linguistic utterances. (shrink)

I advocate a theory of syntactic semantics as a way of understanding how computers can think (and how the Chinese-Room-Argument objection to the Turing Test can be overcome): (1) Semantics, considered as the study of relations between symbols and meanings, can be turned into syntax – a study of relations among symbols (including meanings) – and hence syntax (i.e., symbol manipulation) can suffice for the semantical enterprise (contra Searle). (2) Semantics, considered as the process of understanding one domain (...) (by modeling it) in terms of another, can be viewed recursively: The base case of semantic understanding –understanding a domain in terms of itself – is syntactic understanding. (3) An internal (or narrow ), first-person point of view makes an external (or wide ), third-person point of view otiose for purposes of understanding cognition. (shrink)

This paper is concerned with those elements of planning knowledge that are common to both understanding someone else's plan and creating a plan for one's own use. This planning knowledge can be divided into two bodies: Knowledge about the world, and knowledge about the planning process itself. Our interest here is primarily with the latter corpus. The central thesis is that much of the knowledge about the planning process itself can be formulated in terms of higher‐level goals and plans (...) called meta‐goals and meta‐plans. These entities can then be used by the same understanding and planning mechanisms that process ordinary goals and plans. However, the meta‐planning knowledge now enables these mechanisms to handle much more complicated situations in a uniform manner.Systems based on meta‐planning would have a number of advantages over existing problem solving and understanding systems. The same knowledge could be shared by both a planner and understander, and both would be able to handle complex situations elegantly. In addition, in planning, the use of meta‐planning has several advantages over more traditional methods involving constraints or critics. Meta‐planning allows the full power of a problem solver to be applied to situations that are generally amenable only to special purpose processing. In addition, meta‐planning facilitates the representation of some situations that are difficult to express otherwise. We have begun to introduce meta‐planning knowledge into two systems: PAM, a story understanding program, and PANDORA, a problem solving and plannin̈g system. (shrink)

If we had to indicate in few words the main features of this introductory text to linguistic pragmatics we should maybe begin pointing out the clarity in the exposition. Taking into account its shortness and its pleasant style, we must acknowledge that this book is an excellent introduction to pragmatics. Reading it one not only can make an approach to the central topics and problems in this field, but also can form a clear idea on the wide and varied scope (...) of the discipline. The domain of pragmatics is defined by Green as an intersection of fields within and outside of cognitive science, such as linguistics, cognitive psychology, cultural anthropology and philosophy, and also sociology and rhetoric. But as she remarks. (shrink)

The present paper intends to draw the conception of language implied in the technique of word embeddings that supported the recent development of deep neural network models in computational linguistics. After a preliminary presentation of the basic functioning of elementary artificial neural networks, we introduce the motivations and capabilities of word embeddings through one of its pioneering models, word2vec. To assess the remarkable results of the latter, we inspect the nature of its underlying mechanisms, which have been characterized as (...) the implicit factorization of a word-context matrix. We then discuss the ordinary association of the “distributional hypothesis” with a “use theory of meaning,” often justifying the theoretical basis of word embeddings, and contrast them to the theory of meaning stemming from those mechanisms through the lens of matrix models (such as vector space models and distributional semantic models). Finally, we trace back the principles of their possible consistency through Harris’s original distributionalism up to the structuralist conception of language of Saussure and Hjelmslev. Other than giving access to the technical literature and state of the art in the field of natural language processing to non-specialist readers, the paper seeks to reveal the conceptual and philosophical stakes involved in the recent application of new neural network techniques to the computational treatment of language. (shrink)

We introduce a computational theory of situated language understanding in which the meaning of words and utterances depends on the physical environment and the goals and plans of communication partners. According to the theory, concepts that ground linguistic meaning are neither internal nor external to language users, but instead span the objective‐subjective boundary. To model the possible interactions between subject and object, the theory relies on the notion of perceived affordances: structured units of interaction that can be (...) used for prediction at multiple levels of abstraction. Language understanding is treated as a process of filtering perceived affordances. The theory accounts for many aspects of the situated nature of human language use and provides a unified solution to a number of demands on any theory of language understanding including conceptual combination, prototypicality effects, and the generative nature of lexical items.To support the theory, we describe an implemented system that understands verbal commands situated in a virtual gaming environment. The implementation uses probabilistic hierarchical plan recognition to generate perceived affordances. The system has been evaluated on its ability to correctly interpret free‐form spontaneous verbal commands recorded from unrehearsed game play between human players. The system is able to “step into the shoes” of human players and correctly respond to a broad range of verbal commands in which linguistic meaning depends on social and physical context. We quantitatively compare the system's predictions in response to direct player commands with the actions taken by human players and show generalization to unseen data across a range of situations and verbal constructions. (shrink)

In this paper a logical interpretation of semantic nets and graph grammars is proposed for modelling natural language understanding and creating language understanding computer systems. An example of parsing a Finnish question by graph grammars and inferring the answer to it by a semantic net representation is provided.

While cognitive linguists have been successful at providing accounts of the stable knowledge structures (conceptual metaphors) that give rise to figurative language, and the conceptual mechanisms that manipulate these knowledge structures (conceptual blending), relatively less effort has been thus far devoted to the nature of the linguistic mechanisms involved in figurative language understanding. This paper presents a theoretical account of figurative language understanding, examining metaphor and metonymy in particular. This account is situated within the Theory (...) of Lexical Concepts and Cognitive Models (LCCM Theory). LCCM Theory (Evans, Cognitive Linguistics 17: 491–534, 2006, How words mean: Lexical concepts, cognitive models and meaning construction, Oxford University Press, 2009b) is a cognitively realistic model of lexical representation and semantic compositionality, providing, it is argued, an account of figurative language which complements the ‘backstage’ cognition perspective of Conceptual Blending Theory. It also integrates the notion of conceptual metaphor within the account provided of figurative language understanding. The paper introduces the key mechanisms involved in figurative language understanding arising from language use. The paper also provides a programmatic account of how conceptual metaphors are integrated with linguistic knowledge in figurative language use. It is argued the present proposals flesh out a key aspect of the conceptual integration perspective promoted by Fauconnier and Turner, with which LCCM Theory is continuous. In part, the paper attempts to advance the prospect of a ‘joined up’ cognitive linguistic account of figurative language understanding. (shrink)

Horst Ruthrof revisits Husserl's phenomenology of language and highlights his late writings as essential to understanding the full range of his ideas. Focusing on the idea of language as imaginable as well as the role of a speech community in constituting it, Ruthrof provides a powerful re-assessment of his methodological phenomenology. From the Logical Investigations to untranslated portions of his Nachlass, Ruthrof charts all the developments and amendments in his theorizations. Instead of emphasising the definition and meaning (...) of words, Husserl's later writings point to the variation produced by a community of speakers in the act of communicating. Essential to this linguistic meaning is the intersubjective character that Ruthrof argues is so emblematic of Husserl's position. Using his concepts of intimation, introjection, reciprocity, and voice further outlines a theory that is intersubjective and communal. Bringing his study up to the present day, Ruthrof discusses metal time travel, the evolution of language, and protosyntax in the context of Husserl's late writings, progressing a comprehensive new phenomenological ontology of language with wide-ranging implications for philosophy, linguistics, and cultural studies. (shrink)

Most theories of semantic competence in natural language implicitly assume the Language of Thought Hypothesis. According to this hypothesis, all human cognition consists in the deployment of a language of thought. This language of thought is supposed to be independent of natural language, yet at the same time, it is supposed to be semantically isomorphic with natural language. Given this assumption, it is easy to answer basic questions regarding semantic competence in (...) natural language. What are semantic properties of natural language, and how did they originate? Natural language inherits the semantic properties of the language of thought it is used to express. What does semantic competence consist in? It consists in knowing how to translate items in one's language of thought into a public medium. ;The starting point of this dissertation is the rejection, by some prominent paradigms in cognitive science, of the view that human cognition consists in the deployment of a language of thought that is, at the same time, independent of natural language, yet semantically isomorphic with it. The dissertation constitutes a fledgling attempt to explore an alternative approach to explaining semantic competence in natural language, an approach that is compatible with this rejection of the Language of Thought Hypothesis. ;Rather than conceive of natural language as a tool for expressing items in a pre-existing, semantically isomorphic, language of thought, I urge that language be conceived of as a norm-governed social practice. On this view, the semantic properties of natural language are determined by the inferential norms governing linguistic performances. In order to explain the origin of its semantic properties, one must explain the origin of the inferential norms that govern language. In order to explain what semantic competence consists in, one must explain how individuals learn to abide by these inferential norms. The bulk of the dissertation is devoted to showing that neither of these explanatory projects requires endorsing the Language of Thought Hypothesis. I do this by sketching the form that such explanations might take, and showing that they are consistent with much current research in ethology, comparative psychology, developmental psychology, and evolutionary biology. (shrink)

Young children's early understanding of emotion was investigated by examining their use of emotion terms such as happy, sad, mud, and cry. Five children's emotion language was examined longitudinally from the age of 2 to 5 years, and as a comparison their reference to pains via such terms as burn, sting, and hurt was also examined. In Phase 1 we confirmed and extended prior findings demonstrating that by 2 years of age terms for the basic emotions of happiness, (...) sadness, anger, and fear are commonly used by children as are terms for such related states as crying and hurting. At this early age children produce such terms to refer to self and to others, and to past and future as well as to present states. Over the years from 2 to 5 children's emotion vocabulary expands, their discussion of hypothetical emotions gets underway, and the complexity of their emotion utterances increases. In Phase 2 our analyses go beyond children's production of emotion terms to analyses of their conception of emotion. We focus especially on when children use emotion terms to refer to subjective experiential states of persons. From their earliest uses of these terms in our data children. (shrink)

This paper describes a prototype implementation of a Labelled Deduction System for natural language interpretation, where interpretation is taken to be the process of understanding a natural language utterance. The implementation models the process of understanding wh-gap dependencies in questions and relative clauses for a fragment of English. The paper is divided in three main sections. In Section 1, we introduce the basic architecture of the system. Section 2 outlines a prototype implementation of wh-binding (...) and indicates its potential for explanation of linguistic phenomena, and in Section 3 we briefly set the model within a large theoretical perspective, comparing it to other type-logical approaches to natural language analysis. (shrink)

Meaning depends on context. This applies in obvious cases like deictics or sarcasm as well as more subtle situations like framing or persuasion. One key aspect of this is the identity of the participants in an interaction. Our interpretation of an utterance shifts based on a variety of factors, including personal history, background knowledge, and our relationship to the source. While obviously an incomplete model of individual differences, demographic factors provide a useful starting point and allow us to capture some (...) of this variance. However, the relevance of specific demographic factors varies between situations—where age might be the key factor in one context, ideology might dominate in another. To address this challenge, we introduce a method for combining demographics and context into situated demographic embeddings—mapping representations into a continuous geometric space appropriate for the given domain, showing the resulting representations to be functional and interpretable. We further demonstrate how to make use of related external data so as to apply this approach in low‐resource situations. Finally, we show how these representations can be incorporated into improve modeling of real‐world natural language understanding tasks, improving model performance and helping with issues of data sparsity. (shrink)

This book develops a theory of enriched meanings for natural language interpretation that uses the concept of monads and related ideas from category theory. The volume is interdisciplinary in nature, and will appeal to graduate students and researchers from a range of disciplines interested in natural language understanding and representation.

Natural language processing is a critical part of the digital transformation. NLP enables user-friendly interactions between machine and human by making computers understand human languages. Intelligent chatbot is an essential application of NLP to allow understanding of users’ utterance and responding in understandable sentences for specific applications simulating human-to-human conversations and interactions for problem solving or Q&As. This research studies emerging technologies for NLP-enabled intelligent chatbot development using a systematic patent analytic approach. Some intelligent text-mining techniques are (...) applied, including document term frequency analysis for key terminology extractions, clustering method for identifying the subdomains, and Latent Dirichlet Allocation for finding the key topics of patent set. This research utilizes the Derwent Innovation database as the main source for global intelligent chatbot patent retrievals. (shrink)

A new comprehensive framework for narrative understanding has been developed. Its centerpiece is a new situational logic calledEpisodic Logic, a knowledge and semantic representation well-adapted to the interpretive and inferential needs of general NLU. The most distinctive features of EL is its natural language-like expressiveness. It allows for generalized quantifiers, lambda abstraction, sentence and predicate modifiers, sentence and predicate reification, intensional predicates, unreliable generalizations, and perhaps most importantly, explicit situational variables linked to arbitrary formulas that describe them. (...) These allow episodes to be explicitly related in terms of part-whole, temporal and causal relations. Episodic logical form is easily computed from surface syntax and lends itself to effective inference. (shrink)

The origins of these essays -- Introduction -- Presupposition -- A projection problem for speaker presupposition -- Language and linguistic competence -- Linguistics and psychology -- Semantics and psychology -- Semantics and semantic competence -- The necessity argument -- Truth, meaning, and understanding -- Truth and meaning in perspective -- Semantics and pragmatics -- Naming and asserting -- The gap between meaning and assertion : why what we literally say often differs from what our words literally mean -- (...) Drawing the line between meaning and implicaturem and relating both to assertion -- Descriptions -- Incomplete definite descriptions -- Donnellan's referential/attributive distinction -- Why incomplete descriptions don't refute Russell's theory of descriptions -- Meaning and use : lessons for legal interpretation -- Interpreting legal texts : what is and what is not special about the law. (shrink)

In the field of ʿUlūm al-Qurʾān, in which the conceptual framework of the science of interpretation is drawn and the main rules used in tafsīr are discussed, independent books have been compiled since early periods. Some of these works stand out as foundational texts because they make important determinations about the nature, function, methodology, and relationship of the science of tafsīr with other Islamic sciences. The masterpiece entitled Fahm al-Qurʾān by al-Khāris al-Muhāsibī, a scholar of sufism, tafsīr, kalām, and hadīth (...) is one of these books. Al-Muḥāsibī was a writer who lived in a period when philological and Sharīʿah sciences gained a systematic structure and schools were born and developed. Muhāsib'i's multifaceted scholarly depth, the requirements of his time, and the characteristics of his work distinguish Fahm al-Qurʾān from other works in the field in terms of its problematic, content, and style. It can be said that Fahm al-Qurʾān consists of two main chapters. In the first chapter, the issues of kalām, meaning, and reason are subjected to a quiddity inquiry. In the second chapter, the linguistic aspect of kalām is emphasized and the content and style of the Qur'ān are discussed under the headings of muhkam- mutashābih, naskh and üslūb al-Qur'ān. In this work, al-Muḥāsibī suggests that before asking How we should understand the Qur'ān?, the reader should ask What exactly is being understood and what is meaning? The first of these questions includes the question Who am I? While the sources of The Sciences of the Qurʾān mostly focus on how to understand the Qurʾān at the epistemological and methodological level, in Fahm al- Qurʾān, the meaning of understanding the Qurʾān is problematized and a theory of reason and meaning that is competent with moral training is built. Thus, the conceptual inquiries of the work extend to the existential philosophy of the author. Despite these privileges, it is surprising that Fahm al-Qurʾān did not gain a prominent place among the main sources of tafsir methodology in later periods. This may be due to the theorization of a theory of meaning in which moral maturation is placed at the center. In this research, while trying to reveal the original value that Muhāsibī added to the field with Fahm al-Qurʾān, the systematic structure of the science of interpretation, its connection with other religious sciences, the relationship between speech adjective and language were made problematic and tried to be resolved in the context of the research hypothesis. In this context, it has been determined that al-Muḥāsibī's Fahm al-Qurʾān draws the conceptual framework of tafsīr methodology in relation to the authentic structure of the sharī'ah sciences at a very early period. The fact that Fahm al-Qurʾān carries important signs regarding the position of tafsīr in the hierarchy of sciences and its relationship with other sharī'ah sciences increases the value of the work in identifying the codes of the traditional paradigm. In the research, in which induction methods were used in text analysis and deduction methods were used in discourse analysis, it was determined that Fahm al-Qurʾān had a fiction that shapes the traditional paradigm and depicts the authentic structure of religious sciences. (shrink)

This book presents a new theory of how children acquire language and discusses its implications for a wide range of topics. It explores the roles of innateness and experience in language acquisition, provides further evidence for the theory of Universal Grammar, and shows how linguistic development in children is a driving force behind language shifts and changes.Charles Yang surveys a wide range of errors in children's language and identifies overlooked patterns. He combines these with work in (...) biological evolution in order to develop a model of language acquisition by which to understand the interaction between children's internal linguistic knowledge and their external linguistic experience. He then presents evidence from his own and others' research in the acquisition of syntax and morphology and data from historical language change to test its validity. The model is the first to make quantitative and cross-linguistic predictions about child language. It may also be deployed as a predictive model of language change which, when the evidence is available, could explain why grammars change in a particular direction at a particular time.Knowledge and Learning in Natural Language is a pioneering work at the centre of current concerns in linguistics and cognitive science. It will interest all those concerned to understand and explain language acquisition, Universal Grammar, and language change. (shrink)

Many scholars often use the terms “metaphors,” “analogies,” and “models” interchangeably and inadvertently overlook the uniqueness of each word. According to recent cognitive studies, the three terms involve distinct cognitive processes using features from a familiar concept and applying them to an abstract, complicated concept. In the field of science and religion, there have been various objects or ideas used as metaphors, analogies, or models to describe the science–religion relationship. Although these heuristic tools provided some understanding of the complex (...) interaction, they failed to address the broad nature of science and religion as well as the multifarious relationship between the two in a sociocultural context. Unlike the previous candidates, the concept of language, including the notions of linguistic worldview, linguistic identity, dialects, power, and bilingualism, offers a unique and comprehensive window through which science, religion, and the relationship between the two are seen with clarity. (shrink)

This paper engages with Johaness Bronkhorst’s recognition of a “correspondence principle” as an underlying assumption of Nāgārjuna’s thought. Bronkhorst believes that this assumption was shared by most Indian thinkers of Nāgārjuna’s day, and that it stimulated a broad and fascinating attempt to cope with Nāgārjuna’s arguments so that the principle of correspondence may be maintained in light of his forceful critique of reality. For Bronkhorst, the principle refers to the relation between the words of a sentence and the realities they (...) are meant to convey. While I accept this basic intuition of correspondence, this paper argues that a finer understanding of the principle can be offered. In light of a set of verses from Nāgārjuna’s Śūnyatāsaptati (45–57), it is maintained that for Nāgārjuna, the deeper level of correspondence involves a structural identity he envisions between understanding and reality. Here Nāgārjuna claims that in order for things to exist, a conceptual definition of their nature must be available; in order for there to be a real world and reliable knowledge, a svabhāva of things must be perceived and accounted for. Svabhāva is thus reflected as a knowable essence. Thus, Nāgārjuna’s arguments attacks the accountability of both concepts and things, a position which leaves us with nothing more than mistaken forms of understanding as the reality of the empty. This markedly metaphysical approach is next analyzed in light of the debate Nāgārjuna conducts with a Nyāya interlocutor in his Vigrahavyāvartanī. The correspondence principle is here used to highlight the metaphysical aspect of the debate and to point out the ontological vision of Nāgārjuna’s theory of emptiness. In the analysis of the Vigrahavyāvartanī it becomes clear that the discussion revolves around a foundational metaphysical deliberation regarding the reality or unreality of svabhāva. In this dispute, Nāgārjuna fails to answer the most crucial point raised by his opponent—what is that he defines as empty? (shrink)

My modest aim in this paper is to prove certain relations between some type of hyper-intensional operators, namely context shifting operators, and compositionality in natural languages. Various authors (e.g. von Fintel & Matthewson 2008; Stalnaker 2014) have argued that context-shifting operators are incompatible with compositionality. In fact, some of them understand Kaplan’s (1989) famous ban on context-shifting operators as a constraint on compositionality. Others, (e.g. Rabern 2013) take contextshifting operators to be compatible with compositionality but, unfortunately, do not provide (...) a proof, or an argument in favor of their position. The aim of this paper is to do precisely that. Additionally, I provide a new proof that compositionality for propositional content (intension) is a proper generalization of compositionality for character (hyper-intensions). (shrink)

Addressing biases in natural-language processing (NLP) systems presents an underappreciated ethical dilemma, which we think underlies recent debates about bias in NLP models. In brief, even if we could eliminate bias from language models or their outputs, we would thereby often withhold descriptively or ethically useful information, despite avoiding perpetuating or amplifying bias. Yet if we do not debias, we can perpetuate or amplify bias, even if we retain relevant descriptively or ethically useful information. Understanding this (...) dilemma provides for a useful way of rethinking the ethics of algorithmic bias in NLP. (shrink)

Successful biomedical data mining and information extraction require a complete picture of biological phenomena such as genes, biological processes, and diseases; as these exist on different levels of granularity. To realize this goal, several freely available heterogeneous databases as well as proprietary structured datasets have to be integrated into a single global customizable scheme. We will present a tool to integrate different biological data sources by mapping them to a proprietary biomedical ontology that has been developed for the purposes of (...) making computers understand medical natural language. (shrink)

Juliette Kennedy’s new book brims with intriguing ideas. I don’t understand all of them, and I’m not convinced that the ones I do understand all fit together, b.

Contemporary natural-language semantics began with the assumption that the meaning of a sentence could be modeled by a single truth condition, or by an entity with a truth-condition. But with the recent explosion of dynamic semantics and pragmatics and of work on non- truth-conditional dimensions of linguistic meaning, we are now in the midst of a shift away from a truth-condition-centric view and toward the idea that a sentence’s meaning must be spelled out in terms of its various (...) roles in conversation. This communicative turn in semantics raises historical questions: Why was truth-conditional semantics dominant in the first place, and why were the phenomena now driving the communicative turn initially ignored or misunderstood by truth-conditional semanticists? I offer a historical answer to both questions. The history of natural-language semantics—springing from the work of Donald Davidson and Richard Montague—began with a methodological toolkit that Frege, Tarski, Carnap, and others had created to better understand artificial languages. For them, the study of linguistic meaning was subservient to other explanatory goals in logic, philosophy, and the foundations of mathematics, and this subservience was reflected in the fact that they idealized away from all aspects of meaning that get in the way of a one-to-one correspondence between sentences and truth-conditions. The truth-conditional beginnings of natural- language semantics are best explained by the fact that, upon turning their attention to the empirical study of natural language, Davidson and Montague adopted the methodological toolkit assembled by Frege, Tarski, and Carnap and, along with it, their idealization away from non-truth-conditional semantic phenomena. But this pivot in explana- tory priorities toward natural language itself rendered the adoption of the truth-conditional idealization inappropriate. Lifting the truth-conditional idealization has forced semanticists to upend the conception of linguistic meaning that was originally embodied in their methodology. (shrink)

As modern deep networks become more complex, and get closer to human‐like capabilities in certain domains, the question arises as to how the representations and decision rules they learn compare to the ones in humans. In this work, we study representations of sentences in one such artificial system for natural language processing. We first present a diagnostic test dataset to examine the degree of abstract composable structure represented. Analyzing performance on these diagnostic tests indicates a lack of systematicity (...) in representations and decision rules, and reveals a set of heuristic strategies. We then investigate the effect of training distribution on learning these heuristic strategies, and we study changes in these representations with various augmentations to the training set. Our results reveal parallels to the analogous representations in people. We find that these systems can learn abstract rules and generalize them to new contexts under certain circumstances—similar to human zero‐shot reasoning. However, we also note some shortcomings in this generalization behavior—similar to human judgment errors like belief bias. Studying these parallels suggests new ways to understand psychological phenomena in humans as well as informs best strategies for building artificial intelligence with human‐like language understanding. (shrink)