This paper addresses the question of whether mental representations are compositional. Several researchers have claimed recently that there are empirical data that show mental representations to be context-sensitive in a way that threatens compositionality. Some have then gone on to claim that connectionist encoding schemes are well suited to accommodate such noncom-positionality. I argue here that the data do not show that mental representations are noncompositional, and that there are significant problems with the suggested interpretations of connectionist encoding schemes.

Expectation-based theories of language comprehension, in particular Surprisal Theory, go a long way in accounting for the behavioral correlates of word-by-word processing difficulty, such as reading times. An open question, however, is in which component of the Event-Related brain Potential signal Surprisal is reflected, and how these electrophysiological correlates relate to behavioral processing indices. Here, we address this question by instantiating an explicit neurocomputational model of incremental, word-by-word language comprehension that produces estimates of the N400 and the P600—the two most (...) salient ERP components for language processing—as well as estimates of “comprehension-centric” Surprisal for each word in a sentence. We derive model predictions for a recent experimental design that directly investigates “world-knowledge”-induced Surprisal. By relating these predictions to both empirical electrophysiological and behavioral results, we establish a close link between Surprisal, as indexed by reading times, and the P600 component of the ERP signal. The resultant model thus offers an integrated neurobehavioral account of processing difficulty in language comprehension. (shrink)

Ten years ago, researchers using event-related brain potentials to study language comprehension were puzzled by what looked like a Semantic Illusion: Semantically anomalous, but structurally well-formed sentences did not affect the N400 component—traditionally taken to reflect semantic integration—but instead produced a P600 effect, which is generally linked to syntactic processing. This finding led to a considerable amount of debate, and a number of complex processing models have been proposed as an explanation. What these models have in common is that they (...) postulate two or more separate processing streams, in order to reconcile the Semantic Illusion and other semantically induced P600 effects with the traditional interpretations of the N400 and the P600. Recently, however, these multi-stream models have been called into question, and a simpler single-stream model has been proposed. According to this alternative model, the N400 component reflects the retrieval of word meaning from semantic memory, and the P600 component indexes the integration of this meaning into the unfolding utterance interpretation. In the present paper, we provide support for this “Retrieval–Integration ” account by instantiating it as a neurocomputational model. This neurocomputational model is the first to successfully simulate the N400 and P600 amplitude in language comprehension, and simulations with this model provide a proof of concept of the single-stream RI account of semantically induced patterns of N400 and P600 modulations. (shrink)

How people learn chunks or associations between adjacent items in sequences was modelled. Two previously successful models of how people learn artificial grammars were contrasted: the CCN, a network version of the competitive chunker of Servan‐Schreiber and Anderson [J. Exp. Psychol.: Learn. Mem. Cogn. 16 (1990) 592], which produces local and compositionally‐structured chunk representations acquired incrementally; and the simple recurrent network (SRN) of Elman [Cogn. Sci. 14 (1990) 179], which acquires distributed representations through error correction. The models' susceptibility to two (...) types of interference was determined: prediction conflicts, in which a given letter can predict two other letters that appear next with an unequal frequency; and retroactive interference, in which the prediction made by a letter changes in the second half of training. The predictions of the models were determined by exploring parameter space and seeing howdensely different regions of the space of possible experimental outcomes were populated by model outcomes. For both types of interference, human data fell squarely in regions characteristic of CCN performance but not characteristic of SRN performance. (shrink)

Since its inception, Shannon's information theory has attracted interest for the study of language and music. Recently, a wide range of converging studies have shown how efficient communication pervades language, from phonetics to syntax. Efficient principles imply that more resources should be assigned to highly informative items. For instance, average information content was shown to be a better predictor of word length than frequency, revisiting the famous Zipf's law. However, in spite of the success of the efficient communication framework in (...) the study of language and speech, very little work has investigated its relevance in the analysis of music. Here, we examine the organization of harmonic information in two large corpora of Western music, one made of MIDI files directly sequenced from scores, and the other made of MIDI recordings of live performances of highly skilled piano players. We show that there is a clear positive relationship between (contextual) information content of harmonic sequences and two essential musical properties, namely duration and loudness: the more unexpected a harmonic event is, the longer and the louder it is. (shrink)

Poverty of the Stimulus Arguments have convinced many linguists and philosophers of language that a domain specific language acquisition device (LAD) is necessary to account for language learning. Here we review empirical evidence that casts doubt on the necessity of this domain specific device. We suggest that more attention needs to be paid to the early stages of language acquisition. Many seemingly innate language-related abilities have to be learned over the course of several months. Further, the language input contains rich (...) stochastic information that can be accessed by domain-general learning mechanisms. Computer simulation has shown how mechanisms that are not domain specific can exploit the information contained in language. We conclude that (i) Poverty of the Stimulus Arguments need to be conceptually clarified and (ii) more empirical research needs to be carried out before we can rule out that data driven general purpose mechanisms can account for language learning. (shrink)

The demonstration of a sequential congruency effect in sequence learning has been offered as evidence for control processes that act to inhibit automatic response tendencies via unconscious conflict monitoring. Here we propose an alternative interpretation of this effect based on the associative learning of chains of sequenced contingencies. This account is supported by simulations with a Simple Recurrent Network, an associative model of sequence learning. We argue that the control- and associative-based accounts differ in their predictions concerning the magnitude of (...) the sequential congruency effect across training. These predictions are tested by reanalysing data from a study by Shanks, Wilkinson, and Channon . The results support the associative learning account which explains the sequential congruency effect without appealing to control processes. (shrink)

This paper reviews four significant advances on the feedforward architecture that has dominated discussions of connectionism. The first involves introducing modularity into networks by employing procedures whereby different networks learn to perform different components of a task, and a Gating Network determines which network is best equiped to respond to a given input. The second consists in the use of recurrent inputs whereby information from a previous cycle of processing is made available on later cycles. The third development involves developing (...) compressed representations of strings in which there is no longer an explicit encoding of the components but where information about the structure of the original string can be recovered and so is present functionally. The final advance entails using connectionist learning procedures not just to change weights in networks but to change the patterns used as inputs to the network. These advances significantly increase the usefulness of connectionist networks for modeling human cognitive performance by, among other things, providing tools for explaining the productivity and systematicity of some mental activities, and developing representations that are sensitive to the content they are to represent. (shrink)

Prelinguistic infants must find a way to isolate meaningful chunks from the continuous streams of speech that they hear. BootLex, a new model which uses distributional cues to build a lexicon, demonstrates how much can be accomplished using this single source of information. This conceptually simple probabilistic algorithm achieves significant segmentation results on various kinds of language corpora - English, Japanese, and Spanish; child- and adult-directed speech, and written texts; and several variations in coding structure - and reveals which statistical (...) characteristics of the input have an influence on segmentation performance. BootLex is then compared, quantitatively and qualitatively, with three other groups of computational models of the same infant segmentation process, paying particular attention to functional characteristics of the models and their similarity to human cognition. Commonalities and contrasts among the models are discussed, as well as their implications both for theories of the cognitive problem of segmentation itself, and for the general enterprise of computational cognitive modeling. (shrink)

In agreement with some of the ideas expressed by Perruchet & Vinter (P&V), we believe that some phenomena hitherto attributed to processing may in fact reflect a fundamental distinction between direct and reflexive forms of consciousness. This dichotomy, developed by the phenomenological tradition, is substantiated by examples coming from experimental psychology and lesion neuropsychology.

We identify a number of principles with respect to prediction that, we argue, underpin adult language comprehension: (a) comprehension consists in realizing a mapping between the unfolding sentence and the event representation corresponding to the real‐world event being described; (b) the realization of this mapping manifests as the ability to predict both how the language will unfold, and how the real‐world event would unfold if it were being experienced directly; (c) concurrent linguistic and nonlinguistic inputs, and the prior internal states (...) of the system, each drive the predictive process; (d) the representation of prior internal states across a representational substrate common to the linguistic and nonlinguistic domains enables the predictive process to operate over variable time frames and variable levels of representational abstraction. We review empirical data exemplifying the operation of these principles and discuss the relationship between prediction, event structure, thematic role assignment, and incrementality. (shrink)

How children go about learning the general regularities that govern language, as well as keeping track of the exceptions to them, remains one of the challenging open questions in the cognitive science of language. Computational modeling is an important methodology in research aimed at addressing this issue. We must determine appropriate learning mechanisms that can grasp generalizations from examples of specific usages, and that exhibit patterns of behavior over the course of learning similar to those in children. Early learning of (...) verb argument structure is an area of language acquisition that provides an interesting testbed for such approaches due to the complexity of verb usages. A range of linguistic factors interact in determining the felicitous use of a verb in various constructions—associations between syntactic forms and properties of meaning that form the basis for a number of linguistic and psycholinguistic theories of language. This article presents a computational model for the representation, acquisition, and use of verbs and constructions. The Bayesian framework is founded on a novel view of constructions as a probabilistic association between syntactic and semantic features. The computational experiments reported here demonstrate the feasibility of learning general constructions, and their exceptions, from individual usages of verbs. The behavior of the model over the timecourse of acquisition mimics, in relevant aspects, the stages of learning exhibited by children. Therefore, this proposal sheds light on the possible mechanisms at work in forming linguistic generalizations and maintaining knowledge of exceptions. (shrink)

Despite the considerable attention that the systematicity argument has enjoyed, it is worthwhile examining the argument within the context of similar explanatory arguments from the history of science. This kind of analysis helps show that Connectionism, qua Connectionism, really does not have an explanation of systematicity. Second, and more surprisingly, one finds that the systematicity argument sets such a high explanatory standard that not even Classicism can explain the systematicity of thought.

Despite the considerable attention that the systematicity argument has enjoyed, it is worthwhile examining the argument within the context of similar explanatory arguments from the history of science. This kind of analysis helps show that Connectionism, qua Connectionism, really does not have an explanation of systematicity. Second, and more surprisingly, one finds that the systematicity argument sets such a high explanatory standard that not even Classicism can explain the systematicity of thought.

A basic function of cognition is to detect regularities in sensory input to facilitate the prediction and recognition of future events. It has been proposed that these implicit expectations arise from an internal predictive coding model, based on knowledge acquired through processes such as statistical learning, but it is unclear how different types of statistical information affect listeners’ memory for auditory stimuli. We used a combination of behavioral and computational methods to investigate memory for non-linguistic auditory sequences. Participants repeatedly heard (...) tone sequences varying systematically in their information-theoretic properties. Expectedness ratings of tones were collected during three listening sessions, and a recognition memory test was given after each session. Information-theoretic measures of sequential predictability significantly influenced listeners’ expectedness ratings, and variations in these properties had a significant impact on memory performance. Predictable sequences yielded increasingly better memory performance with increasing exposure. Computational simulations using a probabilistic model of auditory expectation suggest that listeners dynamically formed a new, and increasingly accurate, implicit cognitive model of the information-theoretic structure of the sequences throughout the experimental session. (shrink)

Recent work in the methodology of connectionist explanation has I'ocrrsccl on the notion of levels of explanation. Specific issucs in conncctionisrn hcrc intersect with rvider areas of debate in the philosophy of psychology and thc philosophy of science generally. The issues I raise in this chapter, then, are not unique to cognitive science; but they arise in new and important contexts when connectionism is taken seriously as a model of cognition. The general questions are the relation between levels and the (...) status of levels which have no obvious relation to others. In speaking of levels, what is the connection, if there is one, between explanation and ontology? Which, if any, conccpt of reduction is applicable to connectionist systems? What kind of legitinrtcy can the constructs of common sense psychology, or of that vclsion ol intentional realism represented by classical symbol-systems n I, hirvc irr ir full-scale connectionist theory of mind? (shrink)

The distributional principle according to which morphemes that occur in identical contexts belong, in some sense, to the same category [1] has been advanced as a means for extracting syntactic structures from corpus data. We extend this principle by applying it recursively, and by using mutual information for estimating category coherence. The resulting model learns, in an unsupervised fashion, highly structured, distributed representations of syntactic knowledge from corpora. It also exhibits promising behavior in tasks usually thought to require representations anchored (...) in a grammar, such as systematicity. (shrink)

In 1951, Lashley highlighted the importance of serial order for the brain and behavioural sciences. He considered the response chaining account untenable and proposed an alternative employing parallel response activation and "schemata for action". Subsequently, much has been learned about sequential behaviour, particularly in the linguistic domain. We argue that these developments support Lashley's picture, and recent computational models compatible with it are described. The models are developed in a series of steps, beginning with the basic problem of parallel response (...) competition and its possible resolution into serial action. At each stage, important limitations of the previous models are identified and simple additions proposed to overcome them, including the provision of learning mechanisms. Each type of model is compared with relevant data, and the importance of error data is emphasized. Taken together, the models constitute a unified approach to serial order which has achieved considerable explanatory success across disparate domains. (shrink)

Many activities in Cognitive Science involve complex computer models and simulations of both theoretical and real entities. Artificial Intelligence and the study of artificial neural nets in particular, are seen as major contributors in the quest for understanding the human mind. Computational models serve as objects of experimentation, and results from these virtual experiments are tacitly included in the framework of empirical science. Cognitive functions, like learning to speak, or discovering syntactical structures in language, have been modeled and these models (...) are the basis for many claims about human cognitive capacities. Artificial neural nets (ANNs) have had some successes in the field of Artificial Intelligence, but the results from experiments with simple ANNs may have little value in explaining cognitive functions. The problem seems to be in relating cognitive concepts that belong in the `top-down' approach to models grounded in the `bottom-up' connectionist methodology. Merging the two fundamentally different paradigms within a single model can obfuscate what is really modeled. When the tools (simple artificial neural networks) to solve the problems (explaining aspects of higher cognitive functions) are mismatched, models with little value in terms of explaining functions of the human mind are produced. The ability to learn functions from data-points makes ANNs very attractive analytical tools. These tools can be developed into valuable models, if the data is adequate and a meaningful interpretation of the data is possible. The problem is, that with appropriate data and labels that fit the desired level of description, almost any function can be modeled. It is my argument that small networks offer a universal framework for modeling any conceivable cognitive theory, so that neurological possibility can be demonstrated easily with relatively simple models. However, a model demonstrating the possibility of implementation of a cognitive function using a distributed methodology, does not necessarily add support to any claims or assumptions that the cognitive function in question, is neurologically plausible. (shrink)