Role of prior knowledge in implicit and explicit learning of artificial grammars

Artificial grammar learning (AGL) performance reflects both implicit and explicit processes and has typically been modeled without incorporating any influence from general world knowledge. Our research provides a systematic investigation of the implicit vs. explicit nature of general knowledge and its interaction with knowledge types investigated by past AGL research (i.e., rule- and similarity-based knowledge). In an AGL experiment, a general knowledge manipulation involved expectations being either congruent or incongruent with training stimulus structure. Inconsistent observations paradoxically led to an advantage in structural knowledge and in the use of general world knowledge in both explicit (conscious) and implicit (unconscious) cases (as assessed by subjective measures). The above findings were obtained under conditions of reduced processing time and impaired executive resources. Key findings from our work are that implicit AGL can clearly be affected by general knowledge, and implicit learning can be enhanced by the violation of expectations.     

The role of similarity in artificial grammar learning

The authors examine the role of similarity in artificial grammar learning (AGL; A. S. Reber, 1989). A standard finite-state language was used to create stimuli that were arrangements of embedded geometric shapes (Experiment 1), connected lines (Experiment 2), and sequences of shapes (Experiment 3). Main effects for well-known predictors from the literature (grammaticality, associative global and anchor chunk strength, novel global and anchor chunk strength, length of items, and edit distance) were observed, thus replicating previous work. However, the authors extend previous research by using a widely known similarity-based exemplar model of categorization (the generalized context model; R. M. Nosofsky, 1989) to fit grammaticality judgments, by nested regression analyses. The results suggest that any explanation of AGL that is based on the existing theories is incomplete without a similarity process as well. Also, the results provide a foundation for further interpreting AGL in the wider context of categorization research.     

Theories of artificial grammar learning

Artificial grammar learning (AGL) is one of the most commonly used paradigms for the study of implicit learning and the contrast between rules, similarity, and associative learning. Despite five decades of extensive research, however, a satisfactory theoretical consensus has not been forthcoming. Theoretical accounts of AGL are reviewed, together with relevant human experimental and neuroscience data. The author concludes that satisfactory understanding of AGL requires (a) an understanding of implicit knowledge as knowledge that is not consciously activated at the time of a cognitive operation; this could be because the corresponding representations are impoverished or they cannot be concurrently supported in working memory with other representations or operations, and (b) adopting a frequency-independent view of rule knowledge and contrasting rule knowledge with specific similarity and associative learning (co-occurrence) knowledge.     

Signal-Detection, Threshold, and Dual-Process Models of Recognition Memory: ROCs and Conscious Recollection

Threshold- and signal-detection-based models have dominated theorizing about recognition memory. Building upon these theoretical frameworks, we have argued for a dual-process model in which conscious recollection (a threshold process) and familiarity (a signal-detection process) contribute to memory performance. In the current paper we assessed several memory models by examining the effects of levels of processing and the number of presentations on recognition memory receiver operating characteristics (ROCs). In general, when the ROCs were plotted in probability space they exhibited an inverted U shape; however, when they were plotted inzspace they exhibited a U shape. An examination of the ROCs showed that the dual-process model could account for the observed ROCs, but that models based solely on either threshold or signal-detection processes failed to provide a sufficient account of the data. Furthermore, an examination of subjects' introspective reports using the remember/know procedure showed that subjects were aware of recollection and familiarity and were able to consistently report on their occurrence. The remember/know data were used to accurately predict the shapes of the ROCs, and estimates of recollection and familiarity derived from the ROC data mirrored the subjective reports of these processes.     

Explicit feedback maintains implicit knowledge

The role of feedback was investigated with respect to conscious and unconscious knowledge acquired during artificial grammar learning (AGL). After incidental learning of training sequences, participants classified further sequences in terms of grammaticality and reported their decision strategy with or without explicit veridical feedback. Sequences that disobeyed the learning structure conformed to an alternative structure. Feedback led to an increase in the amount of reported conscious knowledge of structure (derived rules and recollections) but did not increase its accuracy. Conversely, feedback maintained the accuracy of unconscious knowledge of structure (intuition or familiarity-based responses) which otherwise degraded. Results support a dual-process account of AGL. They suggest that implicit learning of the to-be-rejected structure at test contaminates familiarity-based classifications whereas feedback allows competing familiarity signals to be contextualised, which is incompatible with theories that consider familiarity context-insensitive.     

Artificial grammar learning in Alzheimer’s disease

Patients with early Alzheimer’s disease (AD) exhibit impaired declarative memory although some forms of nondeclarative memory are intact. Performance on perceptual nondeclarative memory tasks is often preserved in AD, whereas conceptual nondeclarative memory is often impaired. A conceptual nondeclarative learning task that has been studied in amnesic patients is the artificial grammar learning (AGL) task. Healthy participants and patients with impaired declarative memory both acquire information about an underlying rule structure in this task and exhibit the ability to identify rule-conforming items, despite the subjective experience of guessing at the response. In this study, 12 patients diagnosed with early AD were tested on the AGL task and a matched recognition task. The patients were able to reliably distinguish rule-conforming items from others, indicating successful AGL. Performance of the AD patients was impaired, relative to controls, on a similar recognition task, although they were found to use information about the grammaticality of study items in an attempt to improve their recognition performance. The AD patients showed a dissociation similar to that seen in anterograde amnesia: impaired recognition memory in conjunction with successful AGL. This finding suggests that the brain areas that support AGL are not compromised early in the course of AD. In addition, the nondeclarative memory of the AD patients acquired during AGL appeared to influence their performance on a declarative memory task, suggesting an interaction between this nondeclarative memory task and declarative memory.     

Assertion,knowledge, and action

We argue against the knowledge rule of assertion, and in favour of integrating the account of assertion more tightly with our best theories of evidence and action. We think that the knowledge rule has an incredible consequence when it comes to practical deliberation, that it can be right for a person to do something that she can’t properly assert she can do. We develop some vignettes that show how this is possible, and how odd this consequence is. We then argue that these vignettes point towards alternate rules that tie assertion to sufficient evidence-responsiveness or to proper action. These rules have many of the virtues that are commonly claimed for the knowledge rule, but lack the knowledge rule’s problematic consequences when it comes to assertions about what to do.     

The role of meaning in inflection: why the past tense does not require a rule

How do we produce the past tenses of verbs? For the last 20 years this question has been the focal domain for conflicting theories of language, knowledge representation, and cognitive processing. On one side of the debate have been similarity-based or single-route approaches that propose that all past tenses are formed simply through phonological analogies to existing past tenses stored in memory. On the other side of the debate are rule-based or dual-route approaches which agree that phonological analogy is important for producing irregular past tenses (e.g., think-->thought), but argue that regular past tenses (e.g., walk-->walked) are generated via a +ed rule and that a principled account of regular inflection can only be given by recourse to explicit rules. This debate has become a crucial battleground for arguments concerning the necessity and importance of abstract mental rules, embracing not only language processing, but also the of nature cognition itself. However, in centering on the roles of phonological similarity and rules, the past tense debate has largely ignored the possible role of semantics in determining inflection. This paper presents five studies that demonstrate a striking and decisive role of semantic similarity in inflection. In fact, semantic factors appear to be more important in inflection than the grammatical considerations put forward by the dual-route account. Further, these new findings provide a new way of discriminating between the claims of single-route (similarity-based) and dual-route (rule-based) approaches. It appears that inflection is carried out through analogical reminding based on semantic and phonological similarity and that a rule-based route is not necessary to account for past tense inflection.     

特征的呈现方式对类别学习中规则和相似性知识获得的影响

采用原型变异任务,通过操纵规则和相似性特征的呈现通道,探讨特征的呈现方式对类别学习中规则和相似性知识获得的影响。结果发现,在听觉-视觉条件下习得规则的人数显著多于习得相似性的人数,而在视觉-听觉和视觉-视觉条件下则不存在显著差异;并且,三种条件下习得规则的正确率都高于习得相似性的正确率。这说明特征的呈现方式影响对规则和相似性特征的习得,在听觉通道呈现规则时被试更倾向基于规则分类。     

The contribution of recollection and familiarity to recognition and source-memory judgments: a formal dual-process model and an analysis of receiver operating characteristics

A formal dual-process model that assumes that memory judgments can be based on a threshold recollection process and a signal-detection-based familiarity process is proposed to account for both recognition and source-memory performance. The model was tested in 4 experiments by examining recognition and source-memory receiver operating characteristics (ROCs). In agreement with the predictions of the model, recognition and source memory dissociated in certain conditions. Recognition ROCs were curvilinear in probability space and relatively linear in z-space, as expected if recollection and familiarity contributed to performance. In contrast, source ROCs typically were linear and exhibited a pronounced U shape in z-space, as expected if performance primarily relied on recollection. However, in conditions in which familiarity was clearly indicative of an item's source, the source ROC became curvilinear, suggesting that participants could use familiarity as a basis for source judgments. Several alternative models, including the unequal-variance signal-detection model, were found to be inconsistent with the ROC data.     

Associative and similarity-based processes in categorization decisions

Two experiments were directed at distinguishing associative and similarity-based accounts of systematic differences in categorization time for different items in natural categories. Experiment 1 investigated the correlation of categorization time with three measures of instance centrality in a category. Production frequency (PF), rated typicality, and familiarity from category norms for British participants (Hampton & Gardiner, 1983) were used to predict mean categorization times for 531 words in 12 semantic categories. PF and typicality (but not familiarity) were found to make significant and independent contributions to categorization time. Error rates were related only to typicality (apart from errors made to ambiguous or unknown items). Experiment 2 provided a further dissociation of PF and typicality. Manipulating the difficulty of the task through the relatedness of the false items interacted primarily with the effect of typicality on categorization time, whereas, under conditions of easy discrimination, prior exposure to the category exemplars affected only the contribution of PF to the decision time. The dissociation of typicality and PF measures is interpreted as providing evidence that speeded categorization involves both retrieval of associations indexed by PF and a similarity-based decision process indexed by typicality.     

Is Probabilistic Evidence a Source of Knowledge?

We report a series of experiments examining whether people ascribe knowledge for true beliefs based on probabilistic evidence. Participants were less likely to ascribe knowledge for beliefs based on probabilistic evidence than for beliefs based on perceptual evidence (Experiments 1 and 2A) or testimony providing causal information (Experiment 2B). Denial of knowledge for beliefs based on probabilistic evidence did not arise because participants viewed such beliefs as unjustified, nor because such beliefs leave open the possibility of error. These findings rule out traditional philosophical accounts for why probabilistic evidence does not produce knowledge. The experiments instead suggest that people deny knowledge because they distrust drawing conclusions about an individual based on reasoning about the population to which it belongs, a tendency previously identified by “judgment and decision making” researchers. Consistent with this, participants were more willing to ascribe knowledge for beliefs based on probabilistic evidence that is specific to a particular case (Experiments 3A and 3B).     

Recognition memory ROCs for item and associative information: The contribution of recollection and familiarity

Receiver-operating characteristics (ROCs) were examined in three recognition memory experiments. ROCs for item information (i.e., was this word presented?) were found to be curvilinear. However, ROCs for associative information (i.e., were these two words presented together?) were found to be linear. The results are in agreement with the predictions of a dual-process model that assumes that recognition judgments are -based on familiarity and recollection. Familiarity reflects the assessment of a continuous strength dimension and is well described as a signal detection process, whereas recollection reflects the retrieval of qualitative information about the study episode and behaves like a discrete threshold process. The results showed that memory judgments about items relied on a combination of recollection and familiarity, but that judgments about associations relied primarily on recollection. Further examination of the associative ROCs suggested that subjects were able to recollect that old pairs of items were in the study list, and, under some conditions, that new pairs were not in the study list.     

Variations in recollection: the effects of complexity on source recognition

Whether recollection is a threshold or signal detection process is highly controversial, and the controversy has centered in part on the shape of receiver operating characteristics (ROCs) and z-transformed ROCs (zROCs). U-shaped zROCs observed in tests thought to rely heavily on recollection, such as source memory tests, have provided evidence in favor of the threshold assumption, but zROCs are not always as U-shaped as threshold theory predicts. Source zROCs have been shown to become more linear when the contribution of familiarity to source discriminations is increased, and this may account for the existing results. However, another way in which source zROCs may become more linear is if the recollection threshold begins to break down and recollection becomes more graded and Gaussian. We tested the "graded recollection" account in the current study. We found that increasing stimulus complexity (i.e., changing from single words to sentences) or increasing source complexity (i.e., changing the sources from audio to videos of speakers) resulted in flatter source zROCs. In addition, conditions expected to reduce recollection (i.e., divided attention and amnesia) had comparable effects on source memory in simple and complex conditions, suggesting that differences between simple and complex conditions were due to differences in the nature of recollection, rather than differences in the utility of familiarity. The results suggest that under conditions of high complexity, recollection can appear more graded, and it can produce curved ROCs. The results have implications for measurement models and for current theories of recognition memory.     

What response times tell of children's behavior on the balance scale task

Analysis of accuracy of responses to balance scale problems gives a global idea of the cognitive processes that underlie problem-solving behavior on this task. We show that response times (RTs) provide additional detailed information about the kind and duration of these processes. We derive predictions about the RTs from Siegler's (1981) model for the balance scale task, including the counterintuitive prediction that young adults are slower than children in solving particular balance scale problems. The predictions were tested in a study in which 191 6- to 22-year-old participants were presented with a computerized balance scale task. RTs were analyzed with regression models. In addition to qualitative differences between items, we also modeled quantitative differences between items in the regression models. Analyses supported the predictions and provided additional knowledge on the rules. Rule II was reformulated as a rule that always involves the encoding, but not always the correct application of the distance cue. RTs provided evidence for the use of a buggy-rule and not an addition-rule. Finally, a relation between rule inconsistency and increased RT was found.     

Support for a continuous (single-process) model of recognition memory and source memory

Does memory retrieval occur in a continuous or an all-or-none manner? The shape of the receiver operating characteristic (ROC) has been used to answer this question, with curvilinear and linear memory ROCs indicating continuous and all-or-none retrieval processes, respectively. Signal detection models (e.g., the unequal variance model) correspond to a continuous retrieval process, whereas threshold models (including the multinomial model and the recollection component of the dual-process model) correspond to an all-or-none process. In studies of source memory, Slotnick et al. (2000) and others have observed curvilinear ROCs (supporting the unequal variance model), whereas Yonelinas (1999) observed linear ROCs (supporting the dual-process model). We resolve these seemingly inconsistent results, showing that source memory ROCs are naturally curvilinear but can appear linear when nondiagnostic source information is included in the analysis. Furthermore, the unequal variance model accounted for both recognition memory and source memory ROCs, supporting a continuous process of memory retrieval.     

Negative evidence and inductive generalisation

How do people use past experience to generalise to novel cases? This paper reports four experiments exploring the significance on one class of past experiences: encounters with negative or contrasting cases. In trying to decide whether all ravens are black, what is the effect of learning about a non-raven that is not black? Two experiments with preschool-aged, young school-aged, and adult participants revealed that providing a negative example in addition to a positive example supports generalisation. Two additional experiments went on to ask which kinds of negative examples offer the most support for generalisations. These studies contrasted similarity-based and category-based accounts of inductive generalisation. Results supported category-based predictions, but only for preschool-aged children. Overall, the younger children showed a greater reliance on negative evidence than did older children and adults. Most things we encounter in the world are negative evidence for our generalisations. Understanding the role of negative evidence is central for psychological theories of inductive generalisation.     

AGSuite: Software to conduct feature analysis of artificial grammar learning performance

To simplify the problem of studying how people learn natural language, researchers use the artificial grammar learning (AGL) task. In this task, participants study letter strings constructed according to the rules of an artificial grammar and subsequently attempt to discriminate grammatical from ungrammatical test strings. Although the data from these experiments are usually analyzed by comparing the mean discrimination performance between experimental conditions, this practice discards information about the individual items and participants that could otherwise help uncover the particular features of strings associated with grammaticality judgments. However, feature analysis is tedious to compute, often complicated, and ill-defined in the literature. Moreover, the data violate the assumption of independence underlying standard linear regression models, leading to Type I error inflation. To solve these problems, we present AGSuite, a free Shiny application for researchers studying AGL. The suite’s intuitive Web-based user interface allows researchers to generate strings from a database of published grammars, compute feature measures (e.g., Levenshtein distance) for each letter string, and conduct a feature analysis on the strings using linear mixed effects (LME) analyses. The LME analysis solves the inflation of Type I errors that afflicts more common methods of repeated measures regression analysis. Finally, the software can generate a number of graphical representations of the data to support an accurate interpretation of results. We hope the ease and availability of these tools will encourage researchers to take full advantage of item-level variance in their datasets in the study of AGL. We moreover discuss the broader applicability of the tools for researchers looking to conduct feature analysis in any field.     

ROCs obtained with two signal intensities presented in random order,and a comparison between yes-no and rating ROCs

Two experiments are presented. The first experiment investigated the effect of mixing signals of two different intensities within the same blocks of trials upon receiver operating characteristics (ROCs) obtained with the rating procedure. The ROCs obtained when the signals of differing intensity were mixed in blocks of trials were found to be quite similar to those obtained when only one signal intensity was presented in any given block of trials. In the second experiment ROCs were determined by having subjects adopt different decision criteria on different blocks of trials in a yes-no experiment. The data points obtained in this experiment fell quite close to the ROCs obtained from the same subjects with the rating procedure in the first experiment.