1,721,048 research outputs found
Recommended from our members
Evaluating linguistic knowledge in neural networks
Full text linkWhere does knowledge of language come from? How, for example, do speakers learn the meanings of words or the restrictions on their co-occurrences? This age-old question has age- old answers, from the necessity of direct sensory experience of the world to the existence of an innate language faculty. Recently, neural networks trained on distributional data have proven enormously successful in applied natural language processing tasks, suggesting that they acquire substantial knowledge of language. This dissertation examines what neural networks learn about language. Specifically, I present four studies that characterize the phonological, morphosyntactic and semantic knowledge of neural networks across more than 80 languages. The first of these focuses on phonological features and I show that distributional data of modest size is sufficient to induce human-like phoneme representations using standard neural architectures. The second uses agreement relations as a means of assessing sensitivity to structure dependence in a state-of-the-art model. Using a new cross-linguistic dataset of four types of agreement relations, I demonstrate that the model does capture syntax-sensitive agreement patterns well in general, but I also highlight the specific linguistic contexts in which its performance degrades. The third study looks at the lexical semantics of visual concepts in two domains, comparing neural models to both sighted and blind speakers’ representations. These analyses show that some human-like knowledge is captured, but that the more nuanced structures of the domains are not. Taken together, these first three studies argue that neural networks trained on distributional data are largely accurate yet imperfect models of language. The final study of this dissertation suggests a way forward. In this study, I show that the semantic typology of tense systems is well explained by a domain-general pressure for communicative efficiency and suggest that this same principle is an appropriate inductive bias for neural networks, which may lead to developing more human-like computational models of language
Recommended from our members
Universals and variation in language and thought: Concepts, communication, and semantic structure
Full text linkWhy do languages parcel human experience into categories in the ways they do, and to what extent do these categories in language shape our view of the world? Both language and nonlinguistic cognition vary across cultures, but not arbitrarily, suggesting that there may be universal constraints on how we talk and think. This dissertation explores the sources and consequences of universals and variation in language and thought in four parts.The first study examines a major premise of the universalist view of cognition, that speakers of all languages share a universal conceptual space, which is partitioned by the categories in language. Previous research on color cognition supports this view; when English speakers successively pile-sort colors, their sorting recapitulates an independently proposed hierarchy of color semantics across languages (Boster, 1986). Here I extend that finding to the domain of spatial relations. Levinson et al. (2003) have proposed a hierarchy of spatial category differentiation, and I show that English speakers successively pile-sort spatial scenes in a manner that recapitulates that semantic hierarchy. This finding provides evidence for a specific hierarchy of spatial notions as a model of universals in conceptual structure, and suggests that universal patterns observed across languages reflect general cognitive forces that are available in the minds of speakers of a single language.The second project of this dissertation demonstrates a process by which domain-specific conceptual universals and more general communicative pressures may shape categories in language, extending a previous account (Regier et al., 2015) of semantic universals and variation. In particular, I show that human simulation of cultural transmission in the lab produces systems of semantic categories that converge toward greater informativeness, in the domains of color and spatial relations. These findings suggest that larger-scale cultural transmission over historical time could have produced the diverse yet informative category systems found in the world’s languages. This work supports the communicative efficiency account of semantic universals and variation and establishes a process through which categories in language become increasingly efficient and increasingly universal. The third study extends the previous account of categories in language to cognition more broadly, showing that the same principles that govern efficient semantic systems also characterize nonlinguistic cognition. I provide an account of spatial cognition in which conceptual categories optimize the trade-off between informativeness (making for fine-grained and intuitively organized spatial categories) and simplicity (limiting the number of categories). I find that pile sorts made by speakers of diverse languages match this universal account more closely than they match the semantics of the sorter’s native language. These results suggest that across languages, spatial cognition reflects universal pressures for efficient categorization, and observed universals in category structure and granularity result from these pressures.The final project of this dissertation probes the role of language in online spatial reasoning, using linguistic interference to prevent participants from relying on language in solving a spatial task. In previous work, adult English speakers have been shown to use a spatial frame of reference that differs from that of nonhuman primates and toddlers (Haun et al., 2006), suggesting that learning the spatial frame of reference used in English may motivate a switch away from universal modes of spatial thought. I find that under linguistic interference, despite a sharp increase in error, adult English speakers fail to readopt the spatial frame of reference used by nonhuman primates and toddlers. This finding rules out the possibility that language affects spatial frames of reference online and accordingly argues against Kay and Kempton’s (1984) account, which predicts a removable online role of language. This result raises the stakes of the debate over the role of language in nonlinguistic spatial frames of reference—either something other than language causes alignment between linguistic and nonlinguistic frames of reference, or language learning fundamentally restructures nonlinguistic spatial cognition in a way that is difficult to reverse.The findings of this dissertation in the domain of space, taken together with parallels in other cognitive domains, reinforce an emerging consensus on the relation of language and thought, by which all people share a universal conceptual foundation that may be altered by language. The research here further elaborates this account, suggesting that universals and variation in both language and thought may derive to some extent from general principles of efficiency. At the same time, it challenges the generality of a classic formulation of this view (Kay & Kempton, 1984), motivating future research. In both complementing and challenging an emerging consensus on language and thought, this dissertation informs our view of language, a defining feature of human cognition, and contributes to a more complete understanding of the nature of thought
Recommended from our members
Preliminary Investigations into the Communicative Efficiency of Logographic Writing Systems and Written Languages
Full text linkIt has been argued that language is shaped, at least in part, by functional cognitive pressures to communicate efficiently. The question of whether or not language is efficient has been explored relative to a wide range of linguistic phenomena, and there is mounting evidence that languages do seem to be communicatively efficient. One area which is underexplored in the efficiency literature, however, is whether there is evidence for efficiency pressures shaping writing systems and written language. In the absence of dedicated studies on efficiency in the written modality, it is difficult to say with certainty whether functional cognitive pressures apply to human communication systems across modalities. In addition, because writing systems feature properties which have no direct analogs in other modalities, the efficiency of such properties can only be evaluated through research which focused directly on the efficiency of writing systems and written language.Approached from the direction of the writing systems literature, there have been numerous theoretical proposals over the years regarding what makes a writing system efficient, economical, or practical, either in a vacuum or relative to a particular (type of) spoken language. In addition, it's been proposed, or at least assumed, that certain typological properties of writing systems are inherently impractical, and their presence has been cited as evidence that functional cognitive pressures may exert little, if any, influence over writing systems and written language. However, in both cases, such proposals are rarely followed up with data-driven, computational studies which can quantitatively evaluate their merit.The goal of this document is to help address these gaps. Relative to the efficiency literature, the studies presented here expand the scope to more directly focus on language's written modality. Relative to the writing systems literature, the studies presented here offer a more data-driven approach to concepts which are often theorized, but rarely empirically and quantitatively tested.Chapters 2, 3, and 4 contain a mix of literature review and general background which help contextualize the studies reported in the dissertation's latter chapters. Chapter 2's focus is writing systems and written language: in addition to providing a review of relevant literature and serving as an introductory primer for the reader who is not already familiar with such topics, chapter 2 clarifies the writing systems-related terminology which will be used throughout the dissertation. Chapter 3 starts by covering the theory and methodology behind the concept of efficiency; the chapter then proceeds to a review of the existing literature on how efficiency pressures may shape language, including reviewing the current state of ideas and evidence regarding efficiency in writing systems and written language. Chapter 4 is intended as a short introduction to (written) Sumerian and (written) Japanese, which serve as case studies for the following chapters.Chapters 5 and 6 report on two studies that aim to contribute to the broader question of efficiency pressures in writing systems. In chapter 5, it is asked whether polyvalence (i.e., ambiguity) and logography (i.e., the potential for morphographic and/or semantographic spellings) are used in written Sumerian and written Japanese in ways which facilitate efficient communication, despite being properties which have a somewhat inefficient reputation. By comparing the two target systems against a battery of plausible hypothetical competitor systems, it is found that the real systems do appear to be efficient: very few of the hypothetical competitors would have been both less effortful to use and more informative about sounds or meanings than the real systems. Via the comparison with hypothetical systems, the results of chapter 5 also show that switching to a more transparently phonographic writing system would not have made writing these languages more efficient.Chapter 6 hones in on a more typologically rare property in writing systems, namely the use of semantic determinatives, written units which convey semantic information but do not map on to spoken sounds or morphemes. Using written Sumerian as a case study, it is found that semantic determinatives do seem to have been used efficiently by Sumerian scribes; however, contrary to the most common justification given for the presence of semantic determinatives, the results suggest that determinatives were not used especially efficiently as a means of lexical disambiguation. Rather, while determinatives do sometimes serve to disambiguate otherwise ambiguous spellings, determinatives may have also served as an efficient means of preparing readers for words that occur in less predictable contexts, thereby smoothing out information in Sumerian texts.While preliminary in many respects, the findings presented here support the hypothesis that writing systems and written languages are indeed shaped, at least in part, by functional communicative pressures. Even writing systems which have uneven reputations regarding their utility, such as written Japanese and written Sumerian, appear to be functional and efficient means of communication
Recommended from our members
A Nation under Joint Custody: How Conflicting Family Models divide US Politics
Full text linkAcross the globe and throughout history, politics are regularly divided into "left-leaning" and "right-leaning" camps. Explaining the sources of this conservative-liberal divide has become a major quest in the cognitive and social sciences. Early attempts have focused on self-interest as a possible explanation. However, as the self-interest hypothesis repeatedly failed, researchers' belief in its explanatory power dwindled. Recent investigations have thus begun to tap into genetic material, personality traits, psychological needs, and moral concerns as possible explanations.One early account of the moral underpinnings that guide conservative and liberal politics is Moral Politics Theory (Lakoff, 1996). It proposes that conservatives and liberals endorse different moral worldviews, which conceptually unify their positions on issues as diverse as abortion, education, the economy, or the environment. Those worldviews are grounded in different beliefs about ideal parenting: conservatives endorse a strict-father and liberals a nurturant-parent model. The theory proposes that parenting beliefs are mapped onto national politics via the NATION AS FAMILY metaphor, and that language use in synch with the two models can significantly influence people's perception of political issues. More specifically, Moral Politics Theory has three principal components: First, it holds that the two parenting models form unified and independent belief systems that are predictive of conservatism and liberalism. Second, it predicts that people engage in a metaphoric mapping process when making political judgments, mapping parenting beliefs onto governance. Third, it predicts that issue framings in terms of the two worldviews result in framing effects, and that only those who endorse a given worldview are susceptible to framings that echo it.Despite the academic and political world's fascination with Moral Politics Theory, its three principal components have not been tested to date. The present research is a comprehensive test of Moral Politics Theory's principal components in a series of six studies. Studies 1 and 2 test the two parenting models' internal consistency, conceptual independence and predictive power for conservative and liberal political attitudes. Studies 3 and 4 examine the mediating role of the NATION AS FAMILY metaphor for political judgments in terms of parenting models. Finally, Studies 5 and 6 investigate framing effects associated with the two worldviews
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
- …
