Neural Responses to Structural Incongruencies in Language and ...

Neural Responses to Structural Incongruencies in Language and Statistical Learning Point to Similar Underlying Mechanisms Morten H. Christiansen (mhc27@cornell.edu) Department of Psychology, Cornell University, Ithaca, NY 14853 USA Christopher M. Conway (cmconway@indiana.edu) Department of Psychological & Brain Sciences, Indiana University, Bloomington, IN 47405 USA Luca Onnis (lo35@cornell.edu) Department of Psychology, Cornell University, Ithaca, NY 14853 USA Abstract Christiansen, 2006; Gómez & Gerken, 2000). Statistical learning involves the extraction of regularities and patterns We used event-related potentials (ERPs) to investigate the distribution of brain activity while adults performed (a) a distributed across a set of exemplars in time and/or space, natural language reading task and (b) a statistical learning task typically without direct awareness of what has been learned. involving sequenced stimuli. The same positive ERP Though many researchers assume that statistical learning is deflection, the P600 effect, typically linked to difficult or important for language acquisition and processing (e.g., ungrammatical syntactic processing, was found for structural Gómez & Gerken, 2000), there is very little direct neural incongruencies in both natural language as well as statistical evidence supporting such a claim. There is some evidence learning and had similar topographical distributions. These results suggest that general learning abilities related to the from event-related potential (ERP) studies showing that processing of complex, sequenced material may be implicated structural incongruencies in non-language sequential stimuli in language processing. We conclude that the same neural elicit similar brain responses as those observed for syntactic mechanisms are recruited for both syntactic processing of violations in natural language: a positive shift in the language stimuli and statistical learning of sequential patterns brainwaves observed about 600 msec after the incongruency more generally. known as the P600 effect (Friederici, Steinhauer, & Pfeifer, Keywords: Event-Related Potentials; Statistical Learning; 2002; Lelekov, Dominey, & Garcia-Larrea, 2000; Patel, Language Processing; P600 Gibson, Ratner, Besson, & Holcomb, 1998). Although encouraging, the similarities are inferred across different Introduction subject populations and across different experimental One of the central questions in cognitive science concerns paradigms. Thus, no firm conclusions can be made because the extent to which higher-order cognitive processes in there is no study that provides a direct within-subject humans are either subserved by separate, domain-specific comparison of the ERP responses to both natural language brain mechanisms or whether the same neural substrate may and statistical learning of sequential patterns. support several cognitive functions in a domain-general In this paper, we investigate the possibility that structural fashion. The issue of modularity has played a particularly incongruencies in both natural language and other sequential important role in the study of language, which has stimuli will elicit the same electrophysiological response traditionally been regarded as being strongly modular (e.g., profile, a P600. We provide a within-subject comparison of Friederici, 1995; Pinker, 1991). Given such modular the neural responses to both types of violations, allowing us characterization, the cognitive and neural machinery to directly assess the hypothesis that statistical learning of employed in acquiring and processing language is sequential information is an important cognitive mechanism considered to be uniquely dedicated to language itself. Thus, underlying language processing. Such a demonstration is on this account, little or no overlap in neural substrates important for both theoretical and practical reasons. would be expected between language and other higher-order Statistical learning has become a popular method for cognitive processes. investigating natural language acquisition and processing, Here, we explore the alternative hypothesis that the neural especially in infant populations (e.g., Gómez & Gerken, underpinnings of language may be part of a broader family 2000). Thus, providing direct neural evidence linking of neural mechanisms that the brain recruits when statistical learning to natural language processing is processing sequential information in general. One such type necessary for validating the statistical learning approach to of learning process—employed to encode complex language. Moreover, our study is also of theoretical sequential patterns and also implicated in language importance as it addresses issues relating to the modularity processing—is implicit statistical learning1 (Conway & of language. Before describing our ERP study, we first 1 “Implicit learning” and “statistical learning” have traditionally be touching on the same underlying learning mechanism, which we been studied separately; however, we consider these two terms to hereafter refer to simply as statistical learning. briefly review recent electrophysiological evidence other involving the processing of English sentences. We regarding the neural correlates of both language and hypothesized that overlapping, at least partially but perhaps statistical learning. entirely, neural processes subserve both statistical learning and natural language processing, and thus anticipated ERP Correlates of Language and Statistical obtaining a similar brain response, the P600, to structural Learning incongruencies in both tasks. In ERP studies of syntactic processing, the P600 response Method was originally observed as an increased late positivity recorded around 600 msec after the onset of a word that is Participants syntactically anomalous (e.g., Hagoort, Brown & Eighteen students (17 right-handed; 5 male) from Cornell Groothusen, 1993; Neville, Nicol, Barss, Forster & Garrett University participated in one session and were paid for (1991). P600 responses were also observed at the point of their participation. Data from an additional 4 participants disambiguation in syntactically ambiguous sentences in were excluded because more than 25% of experimental which participants experienced a ‘garden path’ effect (e.g., trials were contaminated due to an excessive number of eye at ‘was’ in ‘The lawyer charged the defendant was lying’). blinks/movements (n=3) or poor data quality (n=1). The age (e.g., Osterhout & Holcomb, 1992). Osterhout & Mobley of the remaining participants ranged between 18 and 22 (1995) found a similar P600 pattern for ungrammatical years (M = 19.8). All were native speakers of English. items in a study of agreement violations in natural language (e.g., ‘The elected officials hope/*hopes to succeed’, and Stimulus Materials ‘The successful woman congratulated herself/*himself’). Statistical learning (SL) task A miniature grammar (see Other violations of long-distance dependencies in natural Figure 1.a)—a slightly simplified version of that used by language have also elicited P600 effects (e.g., Kaan, Harris, Friederici et al. (2002)—was used to produce a set of Gibson, & Holcomb, 2000). Across these studies, the “sentences” consisting of the form subject-verb-object (with typically observed distribution for the P600 is over central object being optional). The grammar specifies four types of and posterior (occipital and parietal) sites. word categories, each with a particular number of tokens The electrophysiological correlates of statistical learning that can comprise it: Noun (Nhave received much less attention. Statistical learning is 1, N2, N3), Verb (V1, V2, V3), Adjective (Aprimarily investigated behaviorally using some sort of 1, A2), and Determiner, the latter containing two subcategories of articles with different distributional variation of the artificial grammar learning (AGL) paradigm properties (d, D). These categories are indicated in Figure (Reber, 1967), in which a finite-state “grammar” is used to 1.a as N, V, A, d, and D, respectively. The grammar generate sequences conforming to arbitrary underlying rules produces sentences composed of nonword tokens, randomly of correct formation. After relatively short exposure to a assigned to the categories for each subject from a set of 10 subset of sequences generated by an artificial grammar, unique tokens: jux, dupp, hep, meep, nib, tam, sig, lum, subjects are able to discriminate between correct and cav, and biff. Each sentence describes a visual scene (i.e., a incorrect sequences with a reasonable degree of accuracy, referent world) consisting of graphical symbols arranged in although they are typically unaware of the constraints that specific ways. For example, each Noun nonword token had govern the sequences. This paradigm has been used to a corresponding shape referent; likewise, each Verb investigate both implicit learning (e.g., Reber, 1967) and nonword token also had a corresponding referent (circle, language acquisition (e.g., Gomez & Gerken, 2000). octagon, square). The Determiner and Adjective tokens did It is possible that the neural processes recruited during not have their own symbols but instead affected the color of artificial grammar learning of sequential stimuli may be at the Noun referents. That is, a Noun preceded by d meant least partly coextensive with neural processes implicated in that the Noun referent would be black; a Noun preceded by natural language (see also Hoen & Dominey, 2000). If this D Ahypothesis holds, it should be possible to find similar neural 1 denoted a green Noun referent while D A2 resulted in a red Noun referent. Note the distributional restriction that d signatures to violations in AGL sequences and natural never occurs with an Adjective whereas D is always language sequences alike. Indeed, Friederici et al. (2002) followed by one. found natural language-like ERP responses from Sixty sentences from the grammar were used for the participants who had learned an artificial language. One of Learning Phase. The nonword form of the sentences these responses, a P600, was also observed for incongruent consisted of written nonword strings (e.g., nib cav jux). musical chord sequences by Patel et al. (1998), who Each nonword string produced from the grammar described detected no statistically significant differences between the a visual scene consisting of the Noun and Verb referents P600 for syntactic and musical structural incongruities. described above. Verb referents always occurred in the These studies suggest that the P600 may reflect the center of the screen. Noun referents appeared either inside operation of a general neural mechanism that handles the Verb referent (for subject Nouns) or outside of the Verb sequential patterns, whether linguistic or not. Therefore, we referent, to the upper right (for object Nouns). An example set out to assess ERP responses in adult subjects on two of a visual scene is shown in Figure 1.b. separate tasks, one involving statistical learning and the liked and when they were ready, they pressed a key to continue. All three Verbs but only the three Nouns preceded by d were included (i.e., only the black Noun referents). The 6 words were presented in random order, 4 times each for a total of 24 trials. In the second Learning sub-phase, the procedure was identical to the first sub-phase but now the other six Noun variations were included, those preceded by D A1 or D A2 (i.e., the red and green Noun referents). The 9 Nouns and 3 Verbs were presented in random order, two times each, for a total of 24 trials. In the third Learning sub-phase, full sentences were presented to participants, with the nonword tokens presented Figure 1: a) The artificial grammar used to generate the adjacent below the corresponding visual scene. The 60 Learning dependency language. The nodes denote word categories and the sentences described above were used for this sub-phase, arrows indicate valid transitions from the beginning node ([) to the each presented in random order, 3 times each. end node (]). b) An example sentence with its associated visual scene (the sequence of word categories below the dashed line is for In the fourth and final Learning sub-phase, participants illustrative purposes only and was not shown to the participants). were again exposed to the same 60 Learning sentences but this time the visual referent scene appeared on its own, prior An additional 30 grammatical sentences were used for the to displaying the corresponding nonword tokens. First, a Test Phase. Thirty ungrammatical sentences were visual scene was shown for 4 sec, and then after a 300 msec additionally used for the Test Phase. To derive violations for pause, the nonword sentences that described the scene were the ungrammatical sentences, tokens of one word category displayed, one word at a time (duration: 350 msec; ISI: 300 in a grammatical sentence were replaced with tokens from a msec). The 60 Learning sentences/scenes were presented in different word category. random order. Natural language (NL) task Two lists, List1 and List2, In the Test Phase, participants were told that they would containing counter-balanced sentence materials were used be presented with new scenes and sentences from the for the natural language task, adapted from Osterhout and artificial language. Half of the sentences would describe the Mobley (1995). Each list consisted of 60 English sentences, scenes according to the same rules of the language as 30 being grammatical and 30 having a violation in terms of before, whereas the other half of the sentences would subject-verb number agreement (e.g., ‘Most cats likes to contain an error with respect to the rules of the language. play outside’). One additional list of 60 sentences was used The participant’s task was to decide which sentences as filler materials, also adapted from Osterhout and Mobley followed the rules correctly and which did not by pressing a (1995). The filler list had 30 grammatical sentences and 30 button on the response pad. The visual referent scenes were sentences that had one of two types of violation: antecedent-presented first, followed by the nonword sentences (with reflexive number (e.g., ‘The Olympic swimmer trained timing identical to Learning sub-phase 4). After the final themselves for the swim meet’) or gender (e.g., ‘The kind word of the sentence was presented, a 1400 msec pause uncle enjoyed herself at Christmas’) agreement. occurred, followed by a test prompt asking for the participant’s response. The 60 Test sentences/scenes were Procedure presented in random order, one time each. Participants were tested individually, sitting in front of a Natural language task Participants were instructed that computer monitor. The participant’s left and right thumbs they would be presented with English sentences appearing were each positioned over the left and right buttons of a on the screen, one word at a time. Their task was to decide button box. All subjects participated in the SL task first and whether each sentence was acceptable or not (by pressing the NL task second. the left or right button), with an unacceptable sentence being one having any type of anomaly and would not be said by a Statistical learning task Participants were instructed that fluent English speaker. Before each sentence, a fixation their job was to learn an artificial “language” consisting of cross was presented for 500 msec in the center of the screen, new words that they would not have seen before and which and then each word of the sentence was presented one at a described different arrangements of visual shapes appearing time for 350 msec, with 300 msec occurring between each on the computer screen. The SL task consisted of two word (thus words were presented with a similar duration and phases, a Learning Phase and a Test Phase, with the ISI as in the SL task). After the final word of the sentence Learning Phase itself consisting of four sub-phases. was presented, a 1400 msec pause occurred followed by a In the first Learning sub-phase, participants were shown a test prompt asking the subject to make a button response Noun or a Verb, one at a time, with the nonword token regarding the sentence’s acceptability. Participants received displayed at the bottom of the screen and its corresponding a total of 120 sentences, 60 from List1 or List2 and 60 from visual referent displayed in the middle of the screen. the Filler list. Participants could observe the scene for as long as they NATURAL LANGUAGE STATISTICAL LEARNING -4µV msec Figure 2: Grand average ERPs elicited for target words for grammatical (dashed) and ungrammatical (solid) continuations in the natural language (left) and statistical learning (right) tasks. The vertical lines mark the onset of the target word. Six electrodes are shown, representative of the left-anterior (25), right-anterior (124), left-central (37), right-central (105), left-posterior (60), and right-posterior (86) regions. Negative voltage is plotted up. (2005): 300-450, 500-700, and 700-900 msec. Separate EEG Recording and Analyses repeated-measures ANOVAs were performed for each The EEG was recorded from 128 scalp sites using the EGI latency window, with grammaticality (grammatical and Geodesic Sensor Net (Tucker, 1993) during the Test Phase ungrammatical), electrode region (anterior, central, and of the SL task and throughout the NL task. All electrode posterior), and hemisphere (left and right) as factors. impedances were kept below 50 kΩ. Recordings were made Geisser-Greenhouse corrections for non-sphericity of with a 0.1 to 100-Hz bandpass filter and digitized at 250 Hz. variance were applied when appropriate. Because the The continuous EEG was segmented into epochs in the description of the results focuses on the effect of the interval -100 msec to +900 msec with respect to the onset of experimental manipulations, effects related to region or the target word that created the structural incongruency. hemisphere are only reported when they interact with Participants were visually shown a display of the real-grammaticality. Results from the omnibus ANOVA are time EEG and observed the effects of blinking, jaw reported first followed by planned comparisons. clenching, and eye movements, and were given specific instructions to avoid or limit such behaviors throughout the Results experiment. Trials with eye-movement artifacts or more Grammaticality Judgments than 10 bad channels were excluded from the average. A Of the test items in the SL task, participants classified channel was considered bad if it reached 200 µV or changed 93.9% correctly. In the NL task, 92.9% of the target more than 100 µV between samples. This resulted in less noun/verb-agreement items were correctly classified. Both than 11% of trials being excluded, evenly distributed across levels of classification were significantly better than chance conditions. ERPs were baseline-corrected with respect to the (p’s < .0001) and not different from one another (p > .5). 100-msec pre-stimulus interval and referenced to an average reference. Separate ERPs were computed for each subject, Event-Related Potentials each condition, and each electrode. Following Barber and Carreiras (2005), six regions of Figure 2 shows the grand average ERP waveforms for interest were defined, each containing the means of 11 grammatical and ungrammatical trials across six electrodes: left anterior (13, 20, 21, 25, 28, 29, 30, 34, 35, representative electrodes (Barber and Carreiras, 2005) for 36, and 40), left central (31, 32, 37, 38, 41, 42, 43, 46, 47, the NL (left) and SL (right) tasks. Visual inspection of the 48, and 50), left posterior (51, 52, 53, 54, 58, 59, 60, 61, 66, ERPs indicates the presence of a left-anterior negativity 67, and 72), right anterior (4, 111, 112, 113, 116, 117, 118, (LAN) in the NL task, but not in the SL task, and a late 119, 122, 123, and 124), right central (81, 88, 94, 99, 102, positivity (P600) at central and posterior sites in both tasks, 103, 104, 105, 106, 109, and 110), and right posterior (77, with a stronger effect in the left-hemisphere and across 78, 79, 80, 85, 86, 87, 92, 93, 97, and 98). posterior regions. These observations were confirmed by the We performed analyses on the mean voltage within the statistical analyses reported below. same three latency windows as in Barber and Carreiras 300-450 msec latency window For the NL data there was a two-way interaction between grammaticality and hemisphere (F(1,17) = 4.71, p < .05). An effect of grammaticality was only found for the left-anterior region, where ungrammatical items were significantly more negative (F(1,17) = 9.52, p < .007), suggesting a LAN. No significant main effects or interactions related to grammaticality were found for the SL data. 500-700 msec latency window There was an overall effect of grammaticality (F(1,17) = 15.96, p < .001) and a significant interaction between grammaticality and region in the NL data (F(2,34) = 8.88, p < .002, ε = .77). This interaction arose due to the differential effect of grammaticality across the anterior and central regions (F(1,17) = 17.55, p < .001). Whereas the negative deflection elicited by the ungrammatical items continued across the left-anterior region (F(1,17) = 5.49, p < .04), a positive -4µV wave was observed for both posterior regions (left: F(1,17) = 15.23, p < .001; right: F(1,17) = 9.40, p < .007) and marginally significant for the left-central region (F(1,17) = msec 3.16, p = .093), indicative of a P600 effect. For the SL data, there was an overall effect of grammaticality (F(1,17) = 13.94, p < .002). A positive Figure 3: Difference waves (ungrammatical minus grammatical) deflection was observed across the left- and right posterior for the language (light-colored) and statistical learning (dark-regions (F(1,17) = 5.74, p < .03; F(1,17) = 4.53, p < .05) colored) tasks. and marginally significant for the left-central region c onducted a repeated-measures analysis between 500 and (F(1,17) = 4.32, p = .053) suggesting a P600 effect similar 700 msec with task as the main factor. to the one elicited by natural language. There was no main effect of task (F(1,17) = .03, p = .87), 700-900 msec latency window A grammaticality × region × nor any significant interactions with region (F(2,34) = 1.47, hemisphere interaction was found (F(2,34) = 3.65, p < .04, ε p = .246, ε = .71) nor hemisphere (F(1,17) = .45, p = .511). = .98) for the NL data, along with a grammaticality × region However, there was a marginal three-way interaction interaction (F(2,34) = 12.66, p < .001, ε = .72) and an (F(2,34) = 2.77, p = .077) but this was due to the differential overall effect of grammaticality (F(1,17) = 9.46, p < .007). modulation of the task and hemisphere factors in the Both interactions were driven by the differential effects of anterior and central regions (F(1,17) = 4.29, p = .054). grammaticality on the ERPs in the anterior and central Indeed, planned comparisons indicated that only in the left-regions (F(1,17) = 21.25, p < .0001), combined with a anterior region was there a significant effect of task due to hemisphere modulation in the three-way interaction (F(1,17) the LAN-associated negative-going difference wave for the = 4.81, p < .05). The negative deflection for ungrammatical language condition (F(1,17) = 4.95, p < .04). No other items continued in the left-anterior region (F(1,17) = 13.93, effects of task were found (F’s < .6). p < .002, as did the positive wave across left- and right-Because LAN has been hypothesized to arise from posterior regions (F(1,17) = 11.70, p < .003; F(1,17) = different neural processes than the P600 (e.g., Friederici, 11.38, p < .004), and which now also emerged over the 1995), our data suggest that the P600 effects we observed in right-central region (F(1,17) = 5.69, p < .03). both tasks are likely to be produced by the same neural A marginal overall effect of grammaticality was found for generators. This suggestion is further supported by a the SL data (F(1,17) = 3.88, p = .065). In this time window regression analysis in which we used the difference between the positive-going deflection had all but disappeared except ungrammatical and grammatical responses averaged across for a marginal effect across the left-central region (F(1,17) = the posterior region for the SL task to predict the mean 4.23, p = .055). difference elicited by the NL task in the same region. The analysis revealed a significant correlation between P600 Comparison of Language and Statistical Learning effects across tasks (R = .50, F(1,16) = 5.34, p < .04): the To more closely compare the ERP responses to structural stronger a participant’s P600 effect was in the SL task, the incongruencies in language and statistical learning, we more pronounced was the corresponding NL P600 in the NL computed ungrammatical-grammatical difference waves for task. each electrode site. Figure 3 shows the resulting waveforms for our six representative electrodes. NL and SL difference waves were compared in the latency range of the P600: we Discussion Acknowledgments This study provided the first direct comparison of This research was supported by Human Frontiers Science electrophysiological brain signatures of statistical learning Program grant RGP0177/2001-B to MHC. and language processing using a within-subject design. The advantage of such a design is that interindividual variance is References held constant, unlike previous studies that compared neural Barber, H. & Carreiras, M. (2005). Grammatical gender and responses between different individuals participating in number agreement in Spanish: An ERP comparison. Journal of different experiments. Following a brief exposure to Cognitive Neuroscience, 17, 137-153. structured sequences in an SL task incorporating visual Conway, C.M. & Christiansen, M.H. (2006). Statistical learning stimuli, our participants showed evidence of having within and between modalities: Pitting abstract against stimulus-implicitly learned the constraints governing the sequences of specific representations. Psychological Science, 17, 905-912. stimuli. Crucially, sequences that contained structural Friederici, A.D. (1995). The time course of syntactic activation incongruencies elicited a P600 signature that was during language processing: A model based on statistically indistinguishable from the P600 elicited by neuropsychological and neurophysiological data. Brain and Language, 50, 259–281. syntactic violations in the NL task. Friederici, A.D., Steinhauer, K., & Pfeifer, E. (2002). Brain One difference between the ERP data from the two tasks signatures of artificial language processing: Evidence was that we observed a LAN for the NL task but not for the challenging the critical period hypothesis. Proceedings of the SL task. The LAN is sometimes observed following National Academy of Sciences, 99, 529-534. syntactic violations and is thought to reflect a relatively Gómez, R.L. & Gerken, L.A. (2000). Infant artificial language automatic parsing process (Friederici et al., 2002). However, learning and language acquisition. Trends in Cognitive Sciences, as in our study, the LAN was absent following both musical 4, 178-186. sequential incongruencies (Patel et al., 1998) and violations Hagoort, P., Brown, C. M., & Groothusen, J. (1993). The Syntactic of a miniature version of Japanese (Mueller, Hahne, Fujii & Positive Shift (SPS) as an ERP measure of syntactic processing. Language and Cognitive Processes, 8, 439-484. Friederici, 2005). One possible explanation is that the LAN Hoen, M. & Dominey, P.F. (2000). ERP analysis of cognitive reflects a truly language-specific neural process; yet, it is sequencing: A left anterior negativity related to structural perhaps more likely that the LAN denotes a response to transformation processing. Neuroreport, 11, 3187-3191. incongruencies in overly-learned patterns, such as language. Kaan, E., Harris, A., Gibson, E., & Holcomb, P. (2000). The P600 Indeed, the results from the Friederici et al. (2002) artificial as an index of syntactic integration difficulty. Language and grammar learning study suggest that with extensive training Cognitive Processes, 15, 159-201. a LAN effect can be obtained. Thus, we suggest that the Lelekov, T., Dominey, P.F., & Garcia-Larrea, L. (2000). lack of a LAN-type effect in our SL task might signal Dissociable ERP profiles for processing rules vs instances in a differences in the two tasks relating to the vastly different cognitive sequencing task. NeuroReport, 11, 1-4. Mueller, J.L., Hahne, A., Fujii, Y. & Friederici, A.D. (2005). amount of experience that our participants had with the Native and nonnative speakers’ processing of a miniature English language versus the patterned stimuli of the SL task. version of Japanese as revealed by ERPs. Journal of Cognitive What is much more certain given our results is that the Neuroscience, 17, 1229-1244. P600 does not appear to be language-specific. That both Neville, H., Nicol, J., Barss, A., Forster, K. I., & Garrett, M. I. tasks elicited the same P600-type signature suggests that the (1991). Syntactically-based sentence processing classes: same overlapping neural mechanisms are involved in both Evidence from event-related brain potentials. Journal of language processing and statistical learning. This validates Cognitive Neuroscience, 3, 151-165. the application of SL paradigms toward the study of Osterhout, L., & Holcomb, P. J. (1992). Event-related brain language acquisition and processing, and suggests that the potentials elicited by syntactic anomaly. Journal of Memory and SL approach will be a fruitful way of studying language. Language, 31, 785-806. Osterhout, L. & Mobley, L.A. (1995). Event-related potentials Finally, our study has important theoretical implications elicited by failure to agree. Journal of Memory and Language, regarding the nature of the neural mechanisms recruited 34, 739-773. during language learning and processing. The results Patel, A.D., Gibson, E., Ratner, J., Besson, M., & Holcomb, P.J. suggest that brain areas responsible for processing words in (1998). Processing syntactic relations in language and music: An sequences are at least partly coextensive with brain areas event-related potential study. Journal of Cognitive responsible for processing other types of complex sequential Neuroscience, 10, 717-733. information such as sequences of sounds, visual objects, or Pinker, S. (1991). Rules of Language, Science, 253, 530-535 events in general. Thus, we conclude that the neural Reber, A.S. (1967). Implicit learning of artificial grammars. Journal of Verbal Learning and Behavior, 6, 855-863. processes recruited for human abilities involving the Tucker, D.M. (1993). Spatial sampling of head electrical fields: encoding, organization, and production of temporally The geodesic sensor net. Electroencephalography and Clinical unfolding events are likely to be shared by processes Neurophysiology, 87, 145-163. typically attributed to language.