Grammatical Strings Research Articles

Neural Basis of the Implicit Learning of Complex Artificial Grammar with Nonadjacent Dependencies.

The capacity for the implicit learning/processing of complex grammar with nonadjacent dependencies is an important feature of human language learning. In this fMRI study, using an implicit AGL paradigm, we explored the neural basis of the implicit learning of the nonadjacent dependency rule, disentangling from sequence-based chunk knowledge (i.e., local sequential regularities or substring) by focusing on the low chunk strength items (which were naturally less similar to training strings), based on tracking neural responses during training and test phases. After listening to and memorizing a series of strings of 10 syllables generated from nonadjacent artificial grammar in the training phase, participants implicitly acquired the knowledge of grammar and chunks. Regarding grammaticality, Broca's area was specifically related to low chunk strength grammatical strings relative to nongrammatical strings in the test phase. This region showed decreased activity with time in the training phase, and a lesser decrease in activity was associated with higher performance in grammar learning. Furthermore, Broca's area showed significantly higher strength of functional connectivity with the left superior temporal gyrus in the low chunk strength grammatical string compared with nongrammatical strings, and this functional connectivity increased with the training time. For the chunks, the performance of accurate discrimination of high chunk strength from low chunk strength nongrammatical strings was predicted by hippocampal activity in the training phase. Converging evidence from the training and test phases showed that Broca's area and its functional connectivity with the left superior temporal gyrus were engaged in the implicit learning/processing of the nonadjacent dependency rule, separating the effects of chunks.

Artificial Grammar Learning in Children With Developmental Language Disorder.

The purpose of the study was to evaluate implicit learning in children with developmental language disorder (DLD) by employing a visual artificial grammar learning task. Thirteen children with DLD and 24 children with typical language development between the ages of 8 and 12 years completed a visual artificial grammar learning task. During the training phase of the task, participants were presented with strings of shapes that followed the underlying structure of a finite grammar. During the testing phase, participants were asked to judge whether new strings were grammatical or nongrammatical. Grammatical judgment of new strings served to measure generalization of the underlying grammatical structure. Endorsement based on chunk strength, or similarity to training exemplars, served to evaluate the extent to which children relied on surface features to guide their task performance. As a group, children with typical development performed better on the artificial grammar learning task, compared with children with DLD, and accepted more grammatical strings regardless of their similarity to training exemplars. Task performance in both groups was not affected by surface features. Performance of children with DLD whose test accuracy exceeded the learning threshold of 0.5 was consistent with a generalization of the underlying grammatical structure that was unaffected by surface features. The study found group differences in learning outcomes between children with and without DLD. Consistent with previous reports, children with typical development correctly endorsed more grammatical strings than children with DLD, suggesting a better acquisition of the grammatical structure. However, there was no evidence to suggest that children in the DLD group (learners and nonlearners) relied on surface features (i.e., familiarity to training exemplars) in their grammatical judgment. These results refute our hypothesis that children in the DLD group would show judgment based on familiarity.

Proposals in meeting minutes

This paper provides an analysis of the lexico-grammatical features used in English and Spanish in the rhetorical part pertaining to the proposal step of a particular genre, meeting minutes. The results have been used in the construction of a computerized writing aid designed to assist Spanish writers when writing minutes in English. Corpus-based contrast reveals that minutes share a common rhetorical structure but show differences in the lexico-grammatical choices associated with the various rhetorical parts. Proposals are conveyed both in English and Spanish by means of lexical verbs, deverbal constructions and certain clause types. Cross-linguistic differences are in the choice of tense, mood and voice. The analysis also suggests different degrees of conventionalization of these resources in the two languages. Results form part of a linguistic prototype that includes three main components: rhetorical information in Spanish, grammatical strings in English and bilingual genre/move restricted glossaries.

Implicit learning is order dependent.

We report two experiments using the artificial-grammar task that demonstrate order dependence in implicit learning. Studying grammatical training strings in different orders did not affect participants' discrimination of grammatical from ungrammatical test strings, but it did affect their judgments about specific test strings. Current accounts of learning in the artificial-grammar task focus on category-level discrimination and largely ignore item-level discrimination. Hence, the results highlight the importance of moving theory from a category- to an item-level of analysis and point to a new way to evaluate and to refine accounts of implicit learning.

Shared Logical Form or Shared Metaphysics? In search of Corpus-derived Empathy in Stylistics

The paper presents a corpus stylistics view on empathy, seen as a mechanism of sharing textual meaning that subsists between authors and their readers. After a brief mention of the relevant developments in psychology and empirical stylistics, the paper focuses on the relevant research in the field of corpus linguistics. The paper shows that the semantic auras of grammar strings are consistent in the language and interact within a text. Thus in a poem, the subtext of the first line, obtainable through the most frequent lexical collocates of its underlying grammar string, may prospect the later developments within the poem by, for example, foreshadowing negativity or even violence. Empathy is thus seen as the sharing of the subliminal layers of meaning which a priori subsist in grammatical strings devoid of lexis, via their most frequent lexical collocates.

Project DyAdd: Implicit learning in adult dyslexia and ADHD

In this study of the project DyAdd, implicit learning was investigated through two paradigms in adults (18-55 years) with dyslexia (n = 36) or with attention deficit/hyperactivity disorder (ADHD, n = 22) and in controls (n = 35). In the serial reaction time (SRT) task, there were no group differences in learning. However, those with ADHD exhibited faster RTs compared to other groups. In the artificial grammar learning (AGL) task, the groups did not differ from each other in their learning (i.e., grammaticality accuracy or similarity choices). Further, all three groups were sensitive to fragment overlap between learning and test-phase items (i.e., similarity choices were above chance). Grammaticality performance of control participants was above chance, but that of participants with dyslexia and participants with ADHD failed to differ from chance, indicating impaired grammaticality learning in these groups. While the main indices of AGL performance, grammaticality accuracy and similarity choices did not correlate with the neuropsychological variables that reflected dyslexia-related (phonological processing, reading, spelling, arithmetic) or ADHD-related characteristics (executive functions, attention), or intelligence, the explicit knowledge for the AGL grammar (i.e., ability to freely generate grammatical strings) correlated positively with the variables of phonological processing and reading. Further, SRT reaction times correlated positively with full scale intelligence quotient (FIQ). We conclude that, in AGL, learning difficulties of the underlying rule structure (as measured by grammaticality) are associated with dyslexia and ADHD. However, learning in AGL is not related to the defining neuropsychological features of dyslexia or ADHD. Instead, the resulting explicit knowledge relates to characteristics of dyslexia.

Bill Louw’s Contextual Prosodic Theory as the basis of (foreign language) classroom corpus stylistics research

Corpus empiricism may alter the act of reading. This began as the reader searched a reference corpus for individual words and phrases. With the admission of lexicographers that intuition no longer suffices in providing a definition, corpus stylistics must go further by showing that a literary text can no longer be properly interpreted if not seen against the background of the wealth of recorded textual experience. This by no means suggests that a literary text may not have a satisfying impact on an individual reader; rather, corpus stylistics enhances our interpretation by means that are easily available. The core of Bill Louw’s stylistic approach is his claim that prior knowledge is no longer perceived as concepts (unsatisfyingly intuitive). Therefore, reference corpora may serve to enhance our stylistic interpretation of a literary text that was clearly written to be appreciated as a unique textual experience. Roughly, a large reference corpus will provide many parallel textual experiences, so that ‘events’ in the studied text are augmented by their counterparts in corpora. Thus, our understanding of the text will be augmented by what is absent from it, but present in the reference corpora. If, furthermore, our classroom is a foreign language one, the reference corpus will serve as missing anguage experience in the foreign language learner, even if the learner is very proficient. After giving a brief overview of Louw’s Contextual Prosodic Theory (CPT) and its implications for classroom corpus stylistics, the paper describes a study conducted with second-year students of English from the University of Belgrade. The aims of the study are to verify Louw’s principle that text reads text and to test the proposed CPT-based methodology. The study consists of a quantitative part (where the learning phase is followed by a final test) and a qualitative part (questionnaire). The proposed methodology relies on confronting the subjects with concordance lines as a means of interpreting a collocation in a given short excerpt, with an absolute minimum of theoretical background. The subjects are tested on semantic prosodies, absent collocates and auras of grammatical strings, through tasks that vary in format. The results obtained are encouraging for CPT, despite the study’s limitations, which are also discussed.

Non-adjacent dependency learning in infants at familial risk of dyslexia

This study tests the hypothesis that developmental dyslexia is (partly) caused by a deficit in implicit sequential learning, by investigating whether infants at familial risk of dyslexia can track non-adjacent dependencies in an artificial language. An implicit learning deficit would hinder detection of such dependencies, which mark grammatical relations (e.g. between 'is' and '-ing' in 'she is happily singing'). In a head-turn experiment with infants aged 1;6, family risk and typically developing infants were exposed to one of two novel languages containing dependencies of the type a-X-c, b-X-d or a-X-d, b-X-c, with fixed first and third elements and twenty-four different X elements. During test, typically developing children listened longer to ungrammatical strings (i.e. that did not correspond to their training language). However, family-risk children did not discriminate between grammatical and ungrammatical strings, indicating deficient implicit learning. The implications of these findings in relation to dyslexia and other language-based disorders are discussed.

The contribution of phonological short-term memory to artificial grammar learning

Three experiments investigated the contribution of phonological short-term memory (STM) to grammar learning by manipulating rehearsal during study of an auditory artificial grammar made up from a vocabulary of spoken Mandarin syllables. Experiment 1 showed that concurrent, irrelevant articulation impaired grammar learning compared with a nonverbal control task. Experiment 2 replicated and extended this finding, showing that repeating the grammatical strings at study improved grammar learning compared with suppressing rehearsal or remaining silent during learning. Experiment 3 found no effects of rehearsal on grammar learning once participants had learned the component syllables. The findings suggest that phonological STM aids artificial grammar learning via effects on vocabulary learning.

Syntactic transfer in artificial grammar learning

In an artificial grammar learning (AGL) experiment, participants were trained with instances of one grammatical structure before completing a test phase in which they were required to discriminate grammatical from randomly created strings. Importantly, the underlying structure used to generate test strings was different from that used to generate the training strings. Despite the fact that grammatical training strings were more similar to nongrammatical test strings than they were to grammatical test strings, this manipulation resulted in a positive transfer effect, as compared with controls trained with nongrammatical strings. It is suggested that training with grammatical strings leads to an appreciation of set variance that aids the detection of grammatical test strings in AGL tasks. The analysis presented demonstrates that it is useful to conceptualize test performance in AGL as a form of unsupervised category learning.

Instruction effects in implicit artificial grammar learning: A preference for grammaticality

Human implicit learning can be investigated with implicit artificial grammar learning, a paradigm that has been proposed as a simple model for aspects of natural language acquisition. In the present study we compared the typical yes–no grammaticality classification, with yes–no preference classification. In the case of preference instruction no reference to the underlying generative mechanism (i.e., grammar) is needed and the subjects are therefore completely uninformed about an underlying structure in the acquisition material. In experiment 1, subjects engaged in a short-term memory task using only grammatical strings without performance feedback for 5 days. As a result of the 5 acquisition days, classification performance was independent of instruction type and both the preference and the grammaticality group acquired relevant knowledge of the underlying generative mechanism to a similar degree. Changing the grammatical stings to random strings in the acquisition material (experiment 2) resulted in classification being driven by local substring familiarity. Contrasting repeated vs. non-repeated preference classification (experiment 3) showed that the effect of local substring familiarity decreases with repeated classification. This was not the case for repeated grammaticality classifications. We conclude that classification performance is largely independent of instruction type and that forced-choice preference classification is equivalent to the typical grammaticality classification.

Grammar or Serial Order?: Discrete Combinatorial Brain Mechanisms Reflected by the Syntactic Mismatch Negativity

If word strings violate grammatical rules, they elicit neurophysiological brain responses commonly attributed to a specifically human language processor or grammar module. However, an ungrammatical string of words is always also a very rare sequence of events and it is, therefore, not always evident whether specifically linguistic processes are at work when neurophysiological grammar indexes are being reported. We here investigate the magnetic mismatch negativity (MNN) to ungrammatical word strings, to very rare grammatical strings, and to common grammatical phrases. In this design, serial order mechanism mapping the sequential probability of words should neurophysiologically dissociate frequent grammatical phrases from both ungrammatical and rare grammatical strings. However, if syntax as a discrete combinatorial system is reflected, the prediction is that the rare, correctly combined items group with the highly frequent grammatical strings and stand out against ungrammatical strings. Using magnetoencephalography as a measure of human brain activity, we replicated the previously reported syntactic mismatch negativity (sMMN), which distinguishes highly unfamiliar ungrammatical word sequences from common grammatical strings. Crucially, a significant interaction demonstrated that the sMMN specifically distinguished syntactic violations from common grammatical strings, but not uncommon from common grammatical word strings. This significant interaction argues in favor of a genuinely grammatical origin of the sMMN and provides direct neurophysiological evidence for a discrete combinatorial system for word and morpheme sequences in the human brain. The data are more difficult to explain in the context of serial order models that map co-occurrence probabilities of words.

Intact artificial grammar learning in patients with cerebellar degeneration and advanced Parkinson’s disease

In an artificial grammar learning task, subjects were asked to memorise short lists of letter strings formed according to complex rules for letter order. After an interval they were unexpectedly asked to discriminate new grammatical strings from strings which used the same letters but violated the sequential constraints of the grammar. Artificial grammar learning can be mastered successfully by amnesic patients and is considered to be an implicit learning task independent of declarative learning and memory mechanisms. In this study, 10 patients with cerebellar degeneration (CD), 21 Parkinson’s disease (PD) and 15 control subjects were tested on artificial grammar learning. Additionally PD patients with advanced disease were examined under adequate medication and dopaminergic withdrawal. All patient groups showed intact artificial grammar learning. Neither cerebellar damage nor basal ganglia dysfunction nor dopaminergic medication impairs or affects artificial grammar learning. Although the patients showed significant executive dysfunction, implicit learning remains intact. The conclusion is that cerebellar and basal ganglia circuits play no essential part in this kind of implicit learning. The results suggest that artificial grammar learning is a cortically mediated function comparable to the mechanism of visual priming.

The role of implicit learning in the acquisition of generative knowledge

This paper reports two experiments using a modified artificial grammar paradigm to study the acquisition of generative knowledge. It employs a starship game and a computerized form of motherese to study acquisition of generative knowledge of a grammar. It was found that initial exposure to a small set of representative cases and an opportunity to organize the cases enhanced ability to generate grammatical strings. Neither seeing a large number of cases nor building a conceptual model of the grammar enhanced acquisition of generative knowledge. However, building a conceptual model did increase learners’ confidence in their ability to generate strings. The quality of strings generated by learners remained fairly constant across attempts, suggesting that their performance reflects true generative knowledge rather than memories of exemplars.

Neural activity differs between explicit and implicit learning of artificial grammar strings: An fMRI study

Functional magnetic resonance imaging was used to investigate the neural areas underlying retrieval of implicit and explicit knowledge about letter strings. Participants studied strings formed according to an artificial grammar, then performed implicit-learning-based judgments (judging the grammatical status of the string) or explicit-learning-based judgments (recognition) on novel grammatical strings. In comparison with a baseline task, recognition and grammatical judgments led to different patterns of neural activation: Recognition activated the right frontal cortex, whereas grammatical judgment activated the left frontal cortex. Recognition led to higher activity in the precuneus and medial occipital cortex, whereas grammatical judgments led to suppression of activity in the precuneus and activation in the lateral occipital cortex. When the surface structure of the strings was changed, grammatical judgments led to bilateral frontal activity and bilateral but left-lateralized activity in the occipital and parietal lobes. These results provide further evidence for a dissociation between the neural bases of implicit and explicit learning.

Realism in confidence judgements of performance based on implicit learning

In two experiments we employed calibration methods to investigate the realism of participants' confidence ratings of their own classification performance based on knowledge acquired after training on an artificial grammar. In Experiment 1 participants showed good realism (but overconfidence) for grammatical strings but very poor realism for non-grammatical strings. Method of training (string repetition in writing or mere exposure) did not affect the realism. Furthermore, the participants underestimated their overall performance. In Experiment 2, using a more complex grammar and controlling for two types of associative chunk-strength, participants showed good realism (but still overconfidence) for both letter and symbol strings, irrespective of grammaticality. Together, these experiments show that implicit learning can give rise to knowledge products that are associated with fairly realistic meta-knowledge. It is argued that both the zero-correlation criterion and the guessing criterion are misplaced when ...

Beyond dissociation logic: evidence for controlled and automatic influences in artificial grammar learning.

Evidence for unconscious learning has typically been based on dissociations between direct and indirect tests of learning. Because of some inherent problems with dissociation logic, we applied the logic of opposition to 2 artificial grammar learning experiments. In Experiment 1, participants were exposed to 2 different sets of letter strings, generated from 2 different grammars, and later rated test strings for grammaticality with either in-concert (rate grammatical strings consistent with either structure) or opposition (rate grammatical only strings from 1 of the structures) instructions. Manipulating response deadline affected controlled, but not automatic influences. In Experiment 2, after similar training, a source-monitoring test was administered from which the in-concert and opposition conditions were derived. The test indicated that varying the retention interval affected controlled, but not automatic, influences. The results are discussed in terms of awareness, knowledge representation, and metacognitive processing.

Artificial grammar learning by 1-year-olds leads to specific and abstract knowledge

Four experiments used the head-turn preference procedure to assess whether infants could extract and remember information from auditory strings produced by a miniature artificial grammar. In all four experiments, infants generalized to new structure by discriminating new grammatical strings from ungrammatical ones after less than 2 min exposure to the grammar. Infants acquired specific information about the grammar as demonstrated by the ability to discriminate new grammatical strings from those with illegal endpoints (Experiment 1). Infants also discriminated new grammatical strings from those with string-internal pairwise violations (Experiments 2 and 3). Infants in Experiment 4 abstracted beyond specific word order as demonstrated by the ability to discriminate new strings produced by their training grammar from strings produced by another grammar despite a change in vocabulary between training and test. We discuss the implications of these findings for the study of language acquisition.

Independent Judgment-Linked and Motor-Linked Forms of Artificial Grammar Learning

Three experiments investigated whether a motor-linked measure (string typing speed) and an judgment-linked measure (grammatical judgment of strings) accessed the same implicit learning mechanisms in the artificial grammar learning task. Participants first studied grammatical strings through observation or through responding to each letter by typing it and then performed typing and grammatical judgment tests. Grammatical judgment test performance was better after observation than after respond learning, whereas typing test performance on higher order relations was worse after observation than after respond learning (Experiment 1). Participants transferred grammatical knowledge across letter sets on the grammatical judgment test, but not on the typing test (Experiment 2). Typing speed did not differ for hits (grammatical strings classified by participants as grammatical) and misses (grammatical strings classified as nongrammatical, Experiment 3). These results are consistent with typing and grammatical judgment tests tapping independent mechanisms and indicate that implicit learning may consist of many different forms of learning rather than being a unitary learning mechanism.

Rule revision with recurrent neural networks

Recurrent neural networks readily process, recognize and generate temporal sequences. By encoding grammatical strings as temporal sequences, recurrent neural networks can be trained to behave like deterministic sequential finite-state automata. Algorithms have been developed for extracting grammatical rules from trained networks. Using a simple method for inserting prior knowledge (or rules) into recurrent neural networks, we show that recurrent neural networks are able to perform rule revision. Rule revision is performed by comparing the inserted rules with the rules in the finite-state automata extracted from trained networks. The results from training a recurrent neural network to recognize a known non-trivial, randomly-generated regular grammar show that not only do the networks preserve correct rules but that they are able to correct through training inserted rules which were initially incorrect (i.e. the rules were not the ones in the randomly generated grammar).