Applying an exemplar model to the artificial-grammar task: Inferring grammaticality from similarity

We present three artificial-grammar experiments. The first used position constraints, and the second used sequential constraints. The third varied both the amount of training and the degree of sequential constraint. Increasing both the amount of training and the redundancy of the grammar benefited participants' ability to infer grammatical status; nevertheless, they were unable to describe the grammar. We applied a multitrace model of memory to the task. The model used a global measure of similarity to assess the grammatical status of the probe and captured performance both in our experiments and in three classic studies from the literature. The model shows that retrieval is sensitive to structure in memory, even when individual exemplars are encoded sparsely. The work ties an understanding of performance in the artificial-grammar task to the principles used to understand performance in episodic-memory tasks.     

Applying an exemplar model to the artificial-grammar task: String completion and performance on individual items

Jamieson and Mewhort (2009a) demonstrated that performance in the artificial-grammar task could be understood using an exemplar model of memory. We reinforce the position by testing the model against data for individual test items both in a standard artificial-grammar experiment and in a string-completion variant of the standard procedure. We argue that retrieval is sensitive to structure in memory. The work ties performance in the artificial-grammar task to principles of explicit memory.     

Applying an exemplar model to the serial reaction-time task: Anticipating from experience

We present a serial reaction time (SRT) task in which participants identified the location of a target by pressing a key mapped to the location. The location of successive targets was determined by the rules of a grammar, and we varied the redundancy of the grammar. Increasing both practice and the redundancy of the grammar reduced response time, but the participants were unable to describe the grammar. Such results are usually discussed as examples of implicit learning. Instead, we treat performance in terms of retrieval from a multitrace memory. In our account, after each trial, participants store a trace comprising the current stimulus, the response associated with it, and the context provided by the immediately preceding response. When a target is presented, it is used as a prompt to retrieve the response mapped to it. As participants practise the task, the redundancy of the series helps point to the correct response and, thereby, speeds retrieval of the response. The model captured performance in the experiment and in classic SRT studies from the literature. Its success shows that the SRT task can be understood in terms of retrieval from memory without implying implicit learning.     

Applying an exemplar model to an implicit rule-learning task: Implicit learning of semantic structure

Studies of implicit learning often examine peoples’ sensitivity to sequential structure. Computational accounts have evolved to reflect this bias. An experiment conducted by Neil and Higham [Neil, G. J., & Higham, P. A.(2012). Implicit learning of conjunctive rule sets: An alternative to artificial grammars. Consciousness and Cognition, 21, 1393–1400] points to limitations in the sequential approach. In the experiment, participants studied words selected according to a conjunctive rule. At test, participants discriminated rule-consistent from rule-violating words but could not verbalize the rule. Although the data elude explanation by sequential models, an exemplar model of implicit learning can explain them. To make the case, we simulate the full pattern of results by incorporating vector representations for the words used in the experiment, derived from the large-scale semantic space models LSA and BEAGLE, into an exemplar model of memory, MINERVA 2. We show that basic memory processes in a classic model of memory capture implicit learning of non-sequential rules, provided that stimuli are appropriately represented.     

Information theory and artificial grammar learning: inferring grammaticality from redundancy

In artificial grammar learning experiments, participants study strings of letters constructed using a grammar and then sort novel grammatical test exemplars from novel ungrammatical ones. The ability to distinguish grammatical from ungrammatical strings is often taken as evidence that the participants have induced the rules of the grammar. We show that judgements of grammaticality are predicted by the local redundancy of the test strings, not by grammaticality itself. The prediction holds in a transfer test in which test strings involve different letters than the training strings. Local redundancy is usually confounded with grammaticality in stimuli widely used in the literature. The confounding explains why the ability to distinguish grammatical from ungrammatical strings has popularized the idea that participants have induced the rules of the grammar, when they have not. We discuss the judgement of grammaticality task in terms of attribute substitution and pattern goodness. When asked to judge grammaticality (an inaccessible attribute), participants answer an easier question about pattern goodness (an accessible attribute).     

Lure similarity affects visual episodic recognition: Detailed tests of a noisy exemplar model

Summed-similarity models of visual episodic recognition memory successfully predict the variation in false alarm rates across different test items. With data averaged across subjects, Kahana and Sekuler demonstrated that subjects' performance appears to change along with the mean similarity among study items; with high interstimulus similarity, subjects were less likely to commit false alarms to similar lures. We examined this effect in detail by systematically varying the coordinates of study and test items along a critical stimulus dimension and measuring memory performance at each point. To reduce uncontrolled variance associated with individual differences in vision, the coordinates of study and test items were scaled according to each subject's discrimination threshold. Fitting each of four summed-similarity models to the individual subjects' data demonstrated a clear superiority for models that take account of interitem similarity on a trialwise basis.     

Generalization and similarity in exemplar models of categorization: Insights from machine learning

Exemplar theories of categorization depend on similarity for explaining subjects’ ability to generalize to new stimuli. A major criticism of exemplar theories concerns their lack of abstraction mechanisms and thus, seemingly, of generalization ability. Here, we use insights from machine learning to demonstrate that exemplar models can actually generalize very well. Kernel methods in machine learning are akin to exemplar models and are very successful in real-world applications. Their generalization performance depends crucially on the chosen similarity measure. Although similarity plays an important role in describing generalization behavior, it is not the only factor that controls generalization performance. In machine learning, kernel methods are often combined with regularization techniques in order to ensure good generalization. These same techniques are easily incorporated in exemplar models. We show that the generalized context model (Nosofsky, 1986) and ALCOVE (Kruschke, 1992) are closely related to a statistical model called kernel logistic regression. We argue that generalization is central to the enterprise of understanding categorization behavior, and we suggest some ways in which insights from machine learning can offer guidance.     

The role of intonation cues in aphasic patients' performance of the grammaticality judgment task

This study investigated the possibility that the reported success of agrammatic aphasic patients in performing auditory grammaticality judgments results from their use of intonational cues to sentence well-formedness. Two agrammatic aphasics, two anomic aphasics, and two nonagrammatic patients with comprehension deficits for semantically reversible sentences were asked to judge grammatical well-formedness in three conditions. Results in a "natural" listening condition replicated the finding of M.C. Linebarger, M.F. Schwartz, and E. M. Saffran (1983, Cognition, 13, 361-392) that agrammatic patients are sensitive to grammaticality despite poor ability to use syntatic cues in comprehension tasks. Two additional listening conditions employed signal-processed auditory stimuli in which the fundamental frequency was normal or was set at a constant level. Right-hemisphere-damaged and normal controls showed no performance deficit for judgments of monotone sentences. The aphasic patients' performance was slightly worse for both signal-processed conditions, but there was little apparent effect of removing sentence intonation on their ability to judge sentence grammaticality. These results indicate that the high levels of performance on grammaticality judgement tasks have not resulted from patients' detection of intonational cues in ungrammatical sentences.     

Towards an exemplar model of face processing: the effects of race and distinctiveness.

Valentine (1991a, 1991b) described a theoretical framework for face recognition in which faces are encoded as locations in a multidimensional space. It was argued that this approach could provide a unified account of the effects of distinctiveness, inversion, and race on face recognition. In this paper we evaluate the ability of this theoretical framework to account for the effects of distinctiveness and race in four experiments in which white British and Japanese faces served as stimuli and both white British and Japanese students acted as subjects. In a recognition memory experiment the expected "own-race bias" was observed as a Race of Subject x Race of Face interaction. Distinctive faces were recognized more accurately than typical faces, but the effect of distinctiveness did not interact with the race of face or the race of subject. Typical faces were classified faster than distinctive faces in a task in which intact faces had to be distinguished from jumbled faces, as found in earlier work, and the effect of distinctiveness did not interact with the race of face or race of subject. In contrast, a task in which subjects classified faces according to their race did show a greater effect of distinctiveness for own-race faces. The results are discussed in relation to the two specific models within the multidimensional space framework identified by Valentine (1991a): a purely exemplar-based model and a norm-based coding model. It is argued that these results are more easily accommodated in terms of a purely exemplar-based model. Some conceptual problems in applying the norm-based coding model to the effect of race are discussed.     

From exemplar to grammar: a probabilistic analogy-based model of language learning

While rules and exemplars are usually viewed as opposites, this paper argues that they form end points of the same distribution. By representing both rules and exemplars as (partial) trees, we can take into account the fluid middle ground between the two extremes. This insight is the starting point for a new theory of language learning that is based on the following idea: If a language learner does not know which phrase-structure trees should be assigned to initial sentences, s/he allows (implicitly) for all possible trees and lets linguistic experience decide which is the "best" tree for each sentence. The best tree is obtained by maximizing "structural analogy" between a sentence and previous sentences, which is formalized by the most probable shortest combination of subtrees from all trees of previous sentences. Corpus-based experiments with this model on the Penn Treebank and the Childes database indicate that it can learn both exemplar-based and rule-based aspects of language, ranging from phrasal verbs to auxiliary fronting. By having learned the syntactic structures of sentences, we have also learned the grammar implicit in these structures, which can in turn be used to produce new sentences. We show that our model mimicks children's language development from item-based constructions to abstract constructions, and that the model can simulate some of the errors made by children in producing complex questions.     

Dative questions: A study in the relation of acceptability to grammaticality of an english sentence type

Generative grammarians have contended that English sentences of the type Who(m) did you give the book? (what are here called ‘dative questions’) are ungrammatical. The incorporation of the necessary restrictions in the grammar of English to account for this, however, requires a weakening of linguistic theory. It would be desirable, therefore, to account for the restriction within performance theory, as has been proposed by Jackendoff and Culicover (1971). Their particular account is shown here to be inadequate. In the course of trying to devise a better account, we found, by two different questionnaire-type experiments, that some English speakers, all from metropolitan New York City, accept dative questions. On the basis of this finding, we theorize that the observed variation in acceptability of dative questions is best accounted for by differences in the perceptual strategies for determining the grammatical relations in perceived clauses that different populations of English listeners use. There are thus no dialect differences, strictly speaking, for dative questions; they are all grammatical for all English speakers.     

Recognizing distinctive faces: A hybrid-similarity exemplar model account

In recognition memory experiments, Nosofsky and Zaki (2003) found that adding discrete distinctive features to continuous-dimension color stimuli helped participants to identify old items as old (the old-item distinctiveness effect), as well as to identify new items as new. The present study tests the extent to which these results generalize to the domain of face recognition. Two experiments were conducted, one using artificial faces and one using natural faces. Artificial faces were used to test memory for faces with discrete distinctive features while controlling the similarity of the faces themselves on more continuous dimensions. The natural-face experiment used the faces of 40 bald men categorized into three groups (typical, isolated, and distinctive) based on experimental ratings of distinctiveness. In both experiments, there were strong effects of the distinctive features on recognition performance. The data were accounted for reasonably well by a hybrid-similarity version of an exemplar recognition model (Nosofsky and Zaki, 2003), which includes a feature-matching mechanism that can provide boosts to an item's self-similarity.     

Tests of an exemplar model for relating perceptual classification and recognition memory

Experiments were conducted in which Ss made classification, recognition, and similarity judgments for 34 schematic faces. A multidimensional scaling (MDS) solution for the faces was derived on the basis of the similarity judgments. This MDS solution was then used in conjunction with an exemplar-similarity model to accurately predict Ss' classification and recognition judgments. Evidence was provided that Ss allocated attention to the psychological dimensions differentially for classification and recognition. The distribution of attention came close to the ideal-observer distribution for classification, and some tendencies in that direction were observed for recognition. Evidence was also provided for interactive effects of individual exemplar frequencies and similarities on classification and recognition, in accord with the predictions of the exemplar model. Unexpectedly, however, the frequency effects appeared to be larger for classification than for recognition.     

New strings for old: The role of similarity processing in an incidental learning task

This paper reports three experiments that examine the role of similarity processing in McGeorge and Burton's (1990) incidental learning task. In the experiments subjects performed a distractor task involving four-digit number strings, all of which conformed to a simple hidden rule. They were then given a forced-choice memory test in which they were presented with pairs of strings and were led to believe that one string of each pair had appeared in the prior learning phase. Although this was not the case, one string of each pair did conform to the hidden rule. Experiment 1 showed that, as in the McGeorge and Burton study, subjects were significantly more likely to select test strings that conformed to the hidden rule. However, additional analyses suggested that rather than having implicitly abstracted the rule, subjects may have been selecting strings that were in some way similar to those seen during the learning phase. Experiments 2 and 3 were designed to try to separate out effects due to similarity from those due to implicit rule abstraction. It was found that the results were more consistent with a similarity-based model than implicit rule abstraction per se.     

Using a paired-associate learning task to assess parent-child phenotypic similarity

The fact of uncertain paternity has led some researchers to hypothesize that children should more closely resemble their fathers than their mothers. The evidence in support of this hypothesis is mixed, partly because the procedures used to measure perceived phenotypic similarity may not be sensitive enough to detect small effects and partly because comparisons are made between fathers and mothers rather than random control groups. In the present experiment the viability of using a paired-associate learning paradigm to investigate parent-child phenotypic similarity is demonstrated using 15 stimulus sets of colored photographs picturing two adults and one child. Using a 2 x 2 between subjects analysis of variance for total errors across learning trials, evidence indicated genetic relatedness for both mothers and fathers influences perceived resemblance and that association of male facial features with those of children, whether paternal or not, are learned more quickly than female facial features. While the significance of genetic relatedness to facial similarity was expected, the overall sex difference was not. However, the additive combination of these variables, genetic relatedness and maleness may explain why children seem to more closely resemble their fathers.     

Multiple regression analyses in artificial-grammar learning: the importance of control groups

In artificial-grammar learning, it is crucial to ensure that above-chance performance in the test stage is due to learning in the training stage but not due to judgemental biases. Here we argue that multiple regression analysis can be successfully combined with the use of control groups to assess whether participants were able to transfer knowledge acquired during training when making judgements about test stimuli. We compared the regression weights of judgements in a transfer condition (training and test strings were constructed by the same grammar but with different letters) with those in a control condition. Predictors were identical in both conditions-judgements of control participants were treated as if they were based on knowledge gained in a standard training stage. The results of this experiment as well as reanalyses of a former study support the usefulness of our approach.     

Effects of age and divided attention on memory components derived for the category exemplar generation task

Age differences in memory performances on a conceptually driven task, the category exemplar generation (CEG) test, were investigated. Thirty-six younger adults and 36 healthy older adults studied word lists in full and divided attention conditions. Recall was tested with category names. The process-dissociation procedure was used to derive estimates of controlled and automatic memory. Old-old adults (70-84 years) exhibited poorer conscious recollection than both younger (18-24) and young-old adults (59-69). In contrast, no age differences were found in estimates of automatic memory. For the younger and older adults, the divided encoding manipulation reduced both the consciously controlled and automatic estimates of memory. The results suggest that the few prior findings of age deficits in priming on the CEG may have been an artifact of contamination from conscious retrieval processes. They also indicate that the opportunity for greater semantic processing enhances the conceptual priming of both younger and older adults.