Sophisticated senator and legislative onion. Whether or not you have ever heard of these things, we all have some intuition that one of them makes much less sense than the other. In this paper, we introduce a large dataset of human judgments about novel adjective-noun phrases. We use these data to test an approach to semantic deviance based on phrase representations derived with compositional distributional semantic methods, that is, methods that derive word meanings from contextual information, and approximate phrase meanings (...) by combining word meanings. We present several simple measures extracted from distributional representations of words and phrases, and we show that they have a significant impact on predicting the acceptability of novel adjective-noun phrases even when a number of alternative measures classically employed in studies of compound processing and bigram plausibility are taken into account. Our results show that the extent to which an attributive adjective alters the distributional representation of the noun is the most significant factor in modeling the distinction between acceptable and deviant phrases. Our study extends current applications of compositional distributional semantic methods to linguistically and cognitively interesting problems, and it offers a new, quantitatively precise approach to the challenge of predicting when humans will find novel linguistic expressions acceptable and when they will not. (shrink)

The recent trend in cognitive robotics experiments on language learning, symbol grounding, and related issues necessarily entails a reduction of sensorimotor aspects from those provided by a human body to those that can be realized in machines, limiting robotic models of symbol grounding in this respect. Here, we argue that there is a need for modeling work in this domain to explicitly take into account the richer human embodiment even for concrete concepts that prima facie relate merely to simple actions, (...) and illustrate this using distributional methods from computational linguistics which allow us to investigate grounding of concepts based on their actual usage. We also argue that these techniques have applications in theories and models of grounding, particularly in machine implementations thereof. Similarly, considering the grounding of concepts in human terms may be of benefit to future work in computational linguistics, in particular in going beyond “grounding” concepts in the textual modality alone. Overall, we highlight the overall potential for a mutually beneficial relationship between the two fields. (shrink)

Abstract Since their inception, distributional models of semantics have been criticized as inadequate cognitive theories of human semantic learning and representation. A principal challenge is that the representations derived by distributional models are purely symbolic and are not grounded in perception and action; this challenge has led many to favor feature-based models of semantic representation. We argue that the amount of perceptual and other semantic information that can be learned from purely distributional statistics has been underappreciated. We compare the representations (...) of three feature-based and nine distributional models using a semantic clustering task. Several distributional models demonstrated semantic clustering comparable with clustering-based on feature-based representations. Furthermore, when trained on child-directed speech, the same distributional models perform as well as sensorimotor-based feature representations of children’s lexical semantic knowledge. These results suggest that, to a large extent, information relevant for extracting semantic categories is redundantly coded in perceptual and linguistic experience. Detailed analyses of the semantic clusters of the feature-based and distributional models also reveal that the models make use of complementary cues to semantic organization from the two data streams. Rather than conceptualizing feature-based and distributional models as competing theories, we argue that future focus should be on understanding the cognitive mechanisms humans use to integrate the two sources. (shrink)

Population counts and longitude and latitude coordinates were estimated for the 50 largest cities in the United States by computational linguistic techniques and by human participants. The mathematical technique Latent Semantic Analysis applied to newspaper texts produced similarity ratings between the 50 cities that allowed for a multidimensional scaling (MDS) of these cities. MDS coordinates correlated with the actual longitude and latitude of these cities, showing that cities that are located together share similar semantic contexts. This finding was replicated using (...) a first‐order co‐occurrence algorithm. The computational estimates of geographical location as well as population were akin to human estimates. These findings show that language encodes geographical information that language users in turn may use in their understanding of language and the world. (shrink)

Previous studies have shown that object properties are processed faster when they follow properties from the same perceptual modality than properties from different modalities. These findings suggest that language activates sensorimotor processes, which, according to those studies, can only be explained by a modal account of cognition. The current paper shows how a statistical linguistic approach of word co-occurrences can also reliably predict the category of perceptual modality a word belongs to (auditory, olfactory–gustatory, visual–haptic), even though the statistical linguistic approach (...) is less precise than the modal approach (auditory, gustatory, haptic, olfactory, visual). Moreover, the statistical linguistic approach is compared with the modal embodied approach in an experiment in which participants verify properties that share or shift modalities. Response times suggest that fast responses can best be explained by the linguistic account, whereas slower responses can best be explained by the embodied account. These results provide further evidence for the theory that conceptual processing is both linguistic and embodied, whereby less precise linguistic processes precede precise simulation processes. (shrink)

Affective information can be retrieved simply by measuring words co‐occurrences in linguistic contexts. Lenci and colleagues demonstrate that the affective measures retrieved from linguistic occurrences predict words’ concreteness: abstract words are more heavily loaded with affective information than concrete ones. These results challenge the Affective grounding hypothesis, suggesting that abstract concepts may be ungrounded and coded only linguistically, and that their affective load may be a linguistic factor.

Time underlies many interesting human behaviors. Thus, the question of how to represent time in connectionist models is very important. One approach is to represent time implicitly by its effects on processing rather than explicitly (as in a spatial representation). The current report develops a proposal along these lines first described by Jordan (1986) which involves the use of recurrent links in order to provide networks with a dynamic memory. In this approach, hidden unit patterns are fed back to themselves: (...) the internal representations which develop thus reflect task demands in the context of prior internal states. A set of simulations is reported which range from relatively simple problems (temporal version of XOR) to discovering syntactic/semantic features for words. The networks are able to learn interesting internal representations which incorporate task demands with memory demands: indeed, in this approach the notion of memory is inextricably bound up with task processing. These representations reveal a rich structure, which allows them to be highly context‐dependent, while also expressing generalizations across classes of items. These representations suggest a method for representing lexical categories and the type/token distinction. (shrink)

Recently, there has been a great deal of interest in the idea that natural language enhances and extends our cognitive capabilities. Supporters of embodied cognition have been particularly interested in the way in which language may provide a solution to the problem of abstract concepts. Toward this end, some have emphasized the way in which language may act as form of cognitive scaffolding and others have emphasized the potential importance of language-based distributional information. This essay defends a version of the (...) cognitive enhancement thesis that integrates and builds on both of these proposals. I argue that the embodied representations associated with language processing serve as a supplementary medium for conceptual processing. The acquisition of a natural language provides a means of extending our cognitive reach by giving us access to an internalized combinatorial symbol system that augments and supports the context-sensitive embodied representational systems that exist independently of language. (shrink)