Neural evidence for a single lexicogrammatical processing system. Jennifer Hughes

Transcription

1 Neural evidence for a single lexicogrammatical processing system Jennifer Hughes j.j.hughes@lancaster.ac.uk

2 Background Approaches to collocation

3 Background Association measures

4 Background EEG, ERPs, and ERP components

5 Background Overview and key findings of earlier PhD Experiments

6 Approaches to collocation Collocation: co-occurrence relation between two words [w]ords are said to collocate with one another if one is more likely to occur in the presence of the other than elsewhere (McEnery and Hardie 2012:240) Approaches differ in terms of whether or not they propose separate systems for lexical and grammatical processing Idiom Principle vs. Open-Choice Principle (Sinclair 1991) Lexical Priming (Hoey 2005) a single system Formulaic language (Wray 2002) holistic vs. analytic system Construction Grammar (e.g. Goldberg 1995) a single system

7 Network model

8 Approaches to collocation The Idiom Principle (Sinclair 1991) - a word becomes associated with a meaning through its repeated occurrence in similar contexts (Sinclair 2004:161) Lexical Priming (Hoey 2005) - collocation is fundamentally a psychological concept (p7) - [a]s a word is acquired through encounters with it in speech and writing, it becomes cumulatively loaded with the contexts and co-texts in which it is encountered (p8)

9 Approaches to collocation Formulaic language (Wray 2002) - quicker access to frequently encountered sequences Construction Grammar - a collocation is a particular instance of a construction - network relationship between construction and collocation and between constructions and other related constructions - [t]he collection of constructions... constitute a highly structured lattice of inter-related information (Goldberg 1995)

10 Association Measures Statistical scores which allow us to distinguish between words which co-occur due to chance, and words which co-occur due to true statistical association (Evert 2008:32). Association Measure Type of Measure Statistic Log-likelihood Pure Significance Mutual information Pure Effect size Z-score Hybrid Effect size and significance T-score Hybrid Frequency and significance Dice coefficient Hybrid Frequency and effect size MI3 Hybrid Frequency and effect size

11 Defining EEG and ERPs Electroencephalography (EEG): a means of measuring electrical potentials in the brain by placing electrodes across the scalp (Harley 2008) Event: experimental stimulus Event-related potentials (ERPs): the momentary changes in electrical activity of the brain when a particular stimulus is presented to a person (Ashcraft and Radvansky 2010)

12 ERP Components ERP component: a scalp-recorded voltage change that reflects a specific neural or psychological process (Kappenman and Luck 2011) N400 lexical/semantic processing (Kutas and Hillyard 1980) P600 syntactic processing (Osterhout and Holcomb 1992)

13 ms... semantic error no error μv... syntactic error (Swaab et al. 2012:422).

14 Two processing systems? The identification of the N400 and the P600 suggests that distinct neurophysiological processes are involved in semantic and syntactic processing, but more recent studies have shown that the N400 can be sensitive to syntactic violations and the P600 can be sensitive to semantic violations (e.g. Geyer et al. 2006)

15 or one processing system? the neural systems supporting syntactic and semantic processing may be linked (Kupperberg et al. 2006) [r]esults such as these raise serious and interesting questions about the relationship between semantic and syntactic processes in the brain (Swaab et al. 2012:26) the interaction between semantic and syntactic processes in the brain may be more dynamic than was previously suggested (Swaab et al. 2012:28).

16 Overview of PhD experiments Experiment 1: A pilot study with native speakers Experiment 2: Native speaker study Experiment 3: Non-native speaker study Experiment 4: Replication and correlation study

17 Key findings in Experiments 1 & 2 Experiment 1: Enlarged anterior-central N400 in non-collocational condition P600 results inconclusive Experiment 2: Enlarged N400 in non-collocational condition, only at midline and right hemisphere electrode sites No P600

18 Aims of Experiment 4 Aim 1: To see whether or not any of the results from Experiments 1 and 2 are replicable. Aim 2: To investigate the strength of the correlation between the forward transition probability of a bigram and N400 amplitude. Aim 3: To find out which association measure most closely correlates with N400 amplitude, and thus may be seen as having the most psychological validity.

19 Method

20 Experimental stimuli TP band Collocational bigrams (forward transition probability in written BNC1994) Non-collocational bigrams 0.8 b<0.9 nineteenth century (0.855) nineteenth position 0.7 b<0.8 prime minister (0.796) prime period 0.6 b<0.7 foreseeable future (0.678) foreseeable weeks 0.5 b<0.6 integral part (0.509) integral thought 0.4 b<0.5 twenty-four hours (0.429) twenty-four patients 0.3 b<0.4 disposable income (0.353) disposable property 0.2 b<0.3 minimum wage (0.246) minimum prize 0.1 b<0.2 vast majority (0.182) vast opportunity 0<b<0.1 crucial point (0.017) crucial night McDonald and Shillcock (2003:648) strong collocations have a mean forward TP of

21 Experimental stimuli Collocational condition - In the foreseeable future the new railway line will be built but the completion date has not yet been confirmed. Non-collocational condition - In the foreseeable weeks the new railway line will be built but the completion date has not yet been confirmed. True/false statement Plans to build a new railway line have been cancelled.

22 Experimental stimuli condition 1 sentences taken from BNC concordance lines all 16 participants exposed to both conditions in one of four counterbalanced lists presented word-by-word at a rate of 500 ms per word (300 ms followed by a 200 ms interstimulus interval) experimental effect time-locked to the second word of the bigram

23 Data analysis Part 1 Mean amplitude (ERPLAB) - between 350 and 500 ms for N400 - between 500 and 650 ms for P600 Repeated measures ANOVA (SPSS) with three factors: Experimental condition (collocational bigrams vs. noncollocational bigrams) Anterior-to-posterior electrode position Left-to-right electrode position

24 Electrode zones

25 Data analysis Part 2 Four step approach to computing a single N400 value for each bigram pair that could be correlated with association measures: 1. Computed a difference wave for each bigram pair 2. Measured the mean amplitude in the ms latency range from each difference wave 3. Extracted the N400 values for the nine representative electrode sites 4. Calculated the mean of the amplitude values from these nine electrode sites Conducted a Pearson correlation, correlating N400 amplitude with forward transition probability, and then with 8 other association measures.

26 Results: Part 1

27 N400 ( ms) mean amplitude is lower in the non-collocational condition (M = 0.146, SD = 2.999) compared to the collocational condition (M = 1.132, SD = 3.353) no main effect - Because the difference between conditions is likely to be large at a subset of the sites and small or even opposite at others, you probably won t see a significant main effect of condition (Luck 2014: 336) significant interaction between condition and anterior-toposterior electrode position: F(2, 820) = 7.28, p =.001

28 Note that the difference between conditions is quantitative rather than qualitative; the waveforms follow the same pattern, with varying amplitudes

29 Collocational condition Non-collocational condition

30 P600 ( ms) mean amplitude is higher in the non-collocational condition (M = 0.754, SD = 7.29) compared to the collocational condition (M = -0,119, SD = 9.2) effect is greatest at central and posterior electrode sites

31 Results: Part 2

32 N400 amplitude and TP

33 Ranking of association measures Association measure Pearson's r p-value 1. Z-score * 2. MI * 3. Dice coefficient * 4. T-score * 5. Backward transition probability Frequency Forward transition probability MI Log-likelihood

34 Key issue raised by the results Why does the N400 in Experiment 4 have the classical scalp distribution, but the N400 in Experiments 1 and 2 do not? Why is a P600 elicited in Experiment 4 but not in Experiments 1 and 2? Mean TP is much higher in Experiment 4 (0.452) than in Experiments 1 and 2 (0.231). Higher TP = stronger collocation = stronger expectation = greater increase in cognitive load when expectation is broken. The N400 and P600 could only be elicited fully when the cognitive load is exceptionally high.

35 Implications and conclusions The difference between conditions is quantitative rather than qualitative, suggesting a scale of collocationality rather than a dichotomy, i.e. one processing system rather than two. Both the N400 and P600 are elicited in response to reading a noncollocation, suggesting that the voltage deflections which are typically known to be associated with lexical/semantic and grammatical processing do not work completely independently instead, they work together as part of a single processing system The two most widely used association measures seem to have the least psychological validity, suggesting that corpus-based studies of lexicogrammar do not always use the optimal association measure for their purposes.

36 References (1) Ashcraft, M. H. and Radvansky, G. A. (2010). Cognition (5th edn.). London: Pearson. Evert, S. (2008). Corpora and collocations. In A. Lüdeling and M. Kytö (Eds.). Corpus Linguistics: An International Handbook. Berlin: Mouton de Gruyter, pp Geyer, A., Holcomb, P., Kuperberg, G., and Pearlmutter, N. (2006). Plausibility and sentence comprehension: An ERP Study. Cognitive Neuroscience Supplement. Goldberg, A. E. (1995). A construction grammar approach to argument structure. Chicago: University of Chicago Press. Harley, T. A. (2008). The psychology of language: From data to theory (3rd edn.). New York: Psychology Press. Hoey, M. (2005). Lexical priming: A new theory of words and language. London: Routledge.

37 References (2) Kappenman, E. S. and Luck, S. J. (2011). ERP components: The ups and downs of brainwave recordings. In E. S. Kappenman and S. J. Luck (Eds.). The Oxford handbook of event related potentials. New York: Oxford University Press, pp Kuperberg, G. R., Caplan, D., Sitnikova, T., Eddy, M., & Holcomb, P. J. (2006). Neural correlates of processing syntactic, semantic, and thematic relationships in sentences. Language and Cognitive Processes, 21: Kutas, M. and Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427): Luck, S. J. (2014). An introduction to the event-related potential technique (2nd edn.). Cambridge, MA: MIT Press. McEnery, T. and Hardie, A. (2012). Corpus linguistics: Methods, theory and practice. Cambridge: Cambridge University Press.

38 References (3) Osterhout, L. and Holcomb, P. (1992). Event-related brain potentials elicited by syntactic anomaly. Language and Cognitive Processes, 8(4): Sinclair, J. (1991). Corpus, concordance, collocation. Oxford: Oxford University Press. Sinclair, J. (2004). Trust the text: Language, corpus, and discourse. London: Routledge. Swaab, T. Y., Ledoux, K., Camblin, C. C. and Boudewyn, M. (2012). Language-related ERP components. In S. J. Luck and E. S. Kappenman (Eds.). The Oxford handbook of event related potentials. New York: Oxford University Press, pp Wray, A. (2002). Formulaic language and the lexicon. Cambridge: Cambridge University Press.