The role of contextual influence in language processing

Introduction

The statement to be discussed is ‘Language processing is predictive’. The concept of ‘predictive’ was developed based on a multilevel model containing both lower-order (e.g., orthographic and phonological) and higher-order (e.g., semantic) linguistic processing. In such a model, researchers attempt to account for language processing based on two main paradigms including (1) a purely bottom-up approach such as modular-based one and (2) an interactive approach with both bottom-up and top-down (predictive) processing. The former such as the dictionary-like Search Model (Forster & Bednall, 1976) hypothesises that lower-order processing should precede and determine higher-order processing. Such a model features the exclusive search of the frequency-ordered linguistic repertoire, which was widely tested based on decontextualised experiments. For example, it is found that, of an intact homograph, the dominant meaning will always be activated without contextual stimuli (Gee, 1997; Simpson & Krueger, 1991). It should be noted that the bottom-up model in this essay can also benefit from contextual feedback such as in Cohort Model (Marslen-Wilson & Warren, 1994; Marslen-Wilson & Welsh, 1978), but the higher-order knowledge could only be activated after the serial recognition of the lower-order linguistic evidence (see McClelland et al., 2006). To compare, the interactive approach argues that higher-order knowledge could also be pre-activated to facilitate word recognition, which would be the focus of the present essay. The evidence supporting the interactive paradigm will be provided in two main aspects including letter-/phoneme-level and word-level, with a discussion of some unsolved issues on this topic.

Letter/Phoneme Level

Two main phenomena will be discussed in letter/phoneme-level processing, that is, the word superiority effect and the Ganong effect.
           The word superiority effect indicates that it would be effortless to identify a letter in word rather than in isolation, showing that lexical knowledge will facilitate the recognition of the letter (Reicher, 1969). In this study (ibid.), after controlling the possible confounding effects of memory, letter recognition was facilitated by meaningful lexicons as higher-order knowledge. This seems to support the interactive approach (McClelland & Rumelhart, 1981) highlighting that pre-lexical mechanism also exists in language processing. However, the true mechanism behind it was questioned based on replication studies demonstrating that factors other than lexical knowledge could contribute to this phenomenon. For instance, after regulating the effects of redundancy (i.e., the known linguistic rule of whether certain letter sequences can appear), the word superiority effect was eliminated with single letter recognition comparable or superior to word-embedded letter (Massaro, 1973; Thompson & Massaro, 1973). Therefore, more neuropsychological approaches were employed to assess this effect; some recent studies using response-time-based evidence (e.g., workload capacity analysis) have shown the superiority effects for letter in words (and pronounceable pseudowords) which does not rely on the existence of masking situation (Houpt et al., 2014). Furthermore, this study (ibid.) also confirmed the bottom-up path of the interactive model by showing that the upside-down and Katakana writing will force the readers to operate the serial checking of each letter.
Similar to the word superiority effect, the Ganong effect is ‘a tendency to make phonetic categorizations that make words’ (Ganong, 1980, p. 110). It also seems to support an interactive model such as the TRACE model (McClelland, 1991) and be against the purely bottom-up model such as Cohort Model (Marslen-Wilson & Warren, 1994; Marslen-Wilson & Welsh, 1978) and Shortlist Model (Norris & McQueen, 2008), because the prior lexical knowledge facilitates the phoneme recognition. For example, eye tracking data shows that an immediate Ganong effect exists even without any target acoustic input, in contrast to the assumption that the Ganong effect is based on activation feedback gradually emerging between the form-meaning link (Kingston et al., 2016). Nevertheless, there exist some bottom-up models which could account for Ganong effect such as Shortlist B based on the Bayesian rule (Norris & McQueen, 2008). In Shortlist B, the phoneme is recognised by computing the maximum likelihood ratio (e.g., the possibility of a word appearing given the existence of other words) which is essentially feedforward rather than prelexical. In this regard, when considering activation feedback, it is argued that, unlike Shortlist B which is optimal and parsimonious, the TRACE model fails to benefit from feedback given noisy input, depends on more presupposed parameters and is inconsistent with some behavioural studies (Norris et al., 2018).
To further examine the difference between the interactive model and Shortlist B, it is useful to separately analyse the frequency-based and lexicality-based Ganong effect. For the frequency-based one, it is argued that the ambiguity between two semantically correct words (such as time and dime) will be interpreted towards the higher-frequent one (Connine et al., 1993), which could be accounted for by both interactive and Bayesian models. Based on this result, Politzer-Ahles et al. (2020) conducted a mixed-factorial research comparing the frequency-based and lexicality-based continua of words on Chinese native speakers, identifying that lexical bias always exerted a significantly larger effect than frequency bias concerning Ganong Effect (b = −1.06, p < 0.001). This study provides evidence of the existence of preactivation of semantic meaning, compared with the frequency-based bottom-up recognition.

Word Level

To further corroborate the arguments above, I will review the studies in word-level processing, the core focus of which will be on semantic priming.
           Semantic priming effects showed that the processing of the lexicon after a flash of its semantically related counterpart is significantly faster than that after an unrelated word (Meyer et al., 1972). As participants will be exposed to the prime for a very short period (e.g., 25ms), the semantic information is argued to be activated automatically rather than strategically. Some studies with other designs (e.g., misspelling word or nonword recognition tasks) also confirm this facilitative effects (O’Connor & Forster, 1981). Based on this phenomenon, visual word processing was hypothesised to be interactive with preactivating the semantically related category of lexicons (see Collins & Loftus, 1975 for Spreading-Activation Theory as one possible account).
However, one core problem here is that the priming effects could be attributed to the associative (e.g., habitual co-occurrence) but not the semantic bond between the prime and the stimulus (Fischler, 1977). The investigations on the associative-semantic difference resulted in inconsistent results, leading to a meta-analysis among 26 studies which confirmed that the automatic semantic priming could exist even without associative relationships (Lucas, 2000). A further microanalysis criticised Lucas’s study and attested that the large effect size reported is the combination of both automatic and strategic priming (Hutchison, 2003). Nevertheless, this later study still confirmed the presence of more semantic than associative priming, especially for the lexicons with similar functional or instrumental relationships. Although much abovementioned lab-based evidence supports the interactive model, the external validity of them is still an issue because the carefully designed stimuli cannot represent daily language use. To address it, the study on ‘naturally occurring degraded speech’ (Mattys & Liss, 2008, p. 1236) also presented that the preactivation of abstract knowledge happens in an early stage without reliance on surface linguistic representations.
Additionally, lexical knowledge other than semantic meaning such as transitivity or gender was also shown to affect word processing in a pre-lexical manner. For example, subjects will be easier to process the stimuli sharing the same type of transitivity (argument structure) with the priming word (Frisson et al., 2005; Salamoura & Williams, 2007a). Concerning grammatical gender, both congruent facilitation and incongruent inhibition were identified in Italian, based on three experiments differing in the explicitness of priming attention (Bates et al., 1996). However, unlike semantic priming, grammar-related priming varies between languages with different syntactic structures (Hohlfeld, 2006). In this respect, bi/multilingual subjects presented a priming effect from the L1 grammatical gender to L2 one (Salamoura & Williams, 2007b). This finding, nevertheless, cannot perfectly support the predictive role of language processing because it is still unclear whether L1-L2 automatic transfer is truly attributed to the pre-activation of lexical knowledge.
Beyond the priming effect, I will then elaborate on the contextual effects in passage reading. In terms of eye movement, if following a bottom-up model, the eye fixation time on each letter should be relatively similar due to the serial processing of the language. However, most eye-tracking-based studies identified ‘saccades’ when lexical knowledge can be accessed from either semantic priming or a previous context. A typical phenomenon is that semantically predictable lexicons (e.g., function words) will often be skipped without serially recognising the orthographic form, supporting the hypothesis that contextual constraint will interact with the visual word recognition (see Altarriba et al., 1996; Rayner et al., 2004). When this prediction-based mechanism exists, an interesting phenomenon is prediction error which highlights the difference between listeners’ expectations and the actual stimuli (Blank & Davis, 2016; Summerfield & Egner, 2009). Blank and Davis (2016) tested Predictive Coding theory and Sharpening theory with functional magnetic resonance imaging (fMRI), with multivariate fMRI simulations that uniquely support the neural mechanism of prediction error. A later fMRI study also shows the activation of pre-lexical knowledge through the 'top-down cortical mechanism’ (Di Liberto et al., 2018, p. 256). In acoustically challenging situations where sensory input was controlled, the interaction between the mobilised left IFG and temporal cortex is still significant which supports a co-activation of bottom-up and top-down mechanism (Bouton et al., 2018).

Some further issues

So far, I have reviewed the evidence for predictive language processing, the limitations of some interactive models, and how researchers attempted to address these limitations. With all the findings cited above, the existence of a top-down path is confirmed to supplement the bottom-up processing. However, the intricate mechanism of language processing is still open to discussion. The first unsolved issue is the difference between visual and spoken language processing because the previous meta-analysis has identified a significantly higher effect of spoken word priming (Cohen’s d = .51) than that of visual word priming (Cohen’s d = .25). If this distinction still exists after regulating the other factors (e.g., the longer echoic trace of spoken word), it would be necessary to investigate the underlying mechanism of activation in these two different modalities.
The second unsolved issue is the role of these two paths in different language processing tasks. Some studies emphasise a more bottom-up processing in clear sensory input and a more top-down mechanism in reduced input clarity (e.g., Blank & Davis, 2016), but how the interaction differs in various scenarios or language processing tasks is still indeterminate. This points to the need to diversify the research settings which benefit the future possible meta-analyses. The third unsolved issue is how the two paths interact in processing. A very recent study based on magnetoencephalography data assessed the role of word frequency and context in the processing (Huizeling et al., 2021). They analysed an open database with 102 subjects reading both the intact and scrambled sentences (Mother of all Unification Studies), leading to a possible more refined model of language processing, that is, the text and context will interact in the ‘late stage’ (> 150 – 250 ms, p.1) while both will also have independent effects in early and late stages alike.

Conclusion

The present article supported an interactive approach to language processing by critically reviewing studies investigating the role of contextual influence. It should be noticed that the facilitative effects of higher-order knowledge were accounted for by both the bottom-up model assuming the additive recognition of words and the interactive model acknowledging the predictive processing with prelexical activation. To argue against the purely bottom-up paradigm, I elaborated on the limitations of some interactive models and how some scholars approached these issues. Although the evidence cited based on different designs and settings supports the interactive model, it is argued that most findings should have been corroborated by further investigations and some suffer from comparatively weak validity. Also, three main unsolved issues in the interactive model were discussed, which appealed to more in-depth and methodically rigorous studies.

1. References

Altarriba, J., Kroll, J. F., Sholl, A., & Rayner, K. (1996). The influence of lexical and conceptual constraints on reading mixed-language sentences: Evidence from eye fixations and naming times. Memory & Cognition, 24(4), 477–492. https://doi.org/10.3758/BF03200936
Bates, E., Devescovi, A., Hernandez, A., & Pizzamiglio, L. (1996). Gender priming in Italian. Perception & Psychophysics, 58(7), 992–1004. https://doi.org/10.3758/BF03206827
Blank, H., & Davis, M. H. (2016). Prediction Errors but Not Sharpened Signals Simulate Multivoxel fMRI Patterns during Speech Perception. PLOS Biology, 14(11), e1002577. https://doi.org/10.1371/journal.pbio.1002577
Bouton, S., Chambon, V., Tyrand, R., Guggisberg, A. G., Seeck, M., Karkar, S., Ville, D. van de, & Giraud, A.-L. (2018). Focal versus distributed temporal cortex activity for speech sound category assignment. Proceedings of the National Academy of Sciences, 115(6), E1299–E1308. https://doi.org/10.1073/pnas.1714279115
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82(6), 407–428. https://doi.org/10.1037/0033-295X.82.6.407
Connine, C. M., Titone, D., & Wang, J. (1993). Auditory word recognition: Extrinsic and intrinsic effects of word frequency. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(1), 81–94. https://doi.org/10.1037/0278-7393.19.1.81
Di Liberto, G. M., Lalor, E. C., & Millman, R. E. (2018). Causal cortical dynamics of a predictive enhancement of speech intelligibility. NeuroImage, 166, 247–258. https://doi.org/10.1016/j.neuroimage.2017.10.066
Fischler, I. (1977). Semantic facilitation without association in a lexical decision task. Memory & Cognition, 5(3), 335–339. https://doi.org/10.3758/BF03197580
Forster, K. I., & Bednall, E. S. (1976). Terminating and exhaustive search in lexical access. Memory & Cognition, 4(1), 53–61. https://doi.org/10.3758/BF03213255
Frisson, S., Rayner, K., & Pickering, M. J. (2005). Effects of Contextual Predictability and Transitional Probability on Eye Movements During Reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(5), 862–877. https://doi.org/10.1037/0278-7393.31.5.862
Ganong, W. F. (1980). Phonetic categorization in auditory word perception. Journal of Experimental Psychology: Human Perception and Performance, 6(1), 110–125. https://doi.org/10.1037/0096-1523.6.1.110
Gee, N. R. (1997). Implicit Memory and Word Ambiguity. Journal of Memory and Language, 36(2), 253–275. https://doi.org/10.1006/jmla.1996.2481
Hohlfeld, A. (2006). Accessing grammatical gender in German: The impact of gender-marking regularities. Applied Psycholinguistics, 27(2), 127–142. https://doi.org/10.1017/S0142716406060218
Houpt, J. W., Townsend, J. T., & Donkin, C. (2014). A new perspective on visual word processing efficiency. Acta Psychologica, 145, 118–127. https://doi.org/10.1016/j.actpsy.2013.10.013
Huizeling, E., Arana, S., Hagoort, P., & Schoffelen, J.-M. (2021). Lexical Frequency and Sentence Context Influence the Brain’s Response to Single Words. Neurobiology of Language, 1–31. https://doi.org/10.1162/nol_a_00054
Hutchison, K. A. (2003). Is semantic priming due to association strength or feature overlap? A microanalytic review. Psychonomic Bulletin & Review, 10(4), 785–813. https://doi.org/10.3758/BF03196544
Kingston, J., Levy, J., Rysling, A., & Staub, A. (2016). Eye movement evidence for an immediate Ganong effect. Journal of Experimental Psychology: Human Perception and Performance, 42(12), 1969–1988. https://doi.org/10.1037/xhp0000269
Lucas, M. (2000). Semantic priming without association: A meta-analytic review. Psychonomic Bulletin & Review, 7(4), 618–630. https://doi.org/10.3758/BF03212999
Marslen-Wilson, W., & Warren, P. (1994). Levels of perceptual representation and process in lexical access: Words, phonemes, and features. Psychological Review, 101(4), 653.
Marslen-Wilson, W., & Welsh, A. (1978). Processing interactions and lexical access during word recognition in continuous speech. Cognitive Psychology, 10(1), 29–63.
Massaro, D. W. (1973). Perception of letters, words, and nonwords. Journal of Experimental Psychology, 100(2), 349–353. https://doi.org/10.1037/h0035681
Mattys, S. L., & Liss, J. M. (2008). On building models of spoken-word recognition: When there is as much to learn from natural “oddities” as artificial normality. Perception & Psychophysics, 70(7), 1235–1242. https://doi.org/10.3758/PP.70.7.1235
McClelland, J. L. (1991). Stochastic interactive processes and the effect of context on perception. Cognitive Psychology, 23(1), 1–44. https://doi.org/10.1016/0010-0285(91)90002-6
McClelland, J. L., Mirman, D., & Holt, L. L. (2006). Are there interactive processes in speech perception? Trends in Cognitive Sciences, 10(8), 363–369. https://doi.org/10.1016/j.tics.2006.06.007
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: I. An account of basic findings. Psychological Review, 88(5), 375.
Meyer, D. E., Schvaneveldt, R. W., & Ruddy, M. G. (1972). Activation of lexical memory. Meeting of the Psychonomic Society, St. Louis.
Norris, D., & McQueen, J. M. (2008). Shortlist B: A Bayesian model of continuous speech recognition. Psychological Review, 115(2), 357.
Norris, D., McQueen, J. M., & Cutler, A. (2018). Commentary on “Interaction in Spoken Word Recognition Models”. Frontiers in Psychology, 9, 1568. https://doi.org/10.3389/fpsyg.2018.01568
O’Connor, R. E., & Forster, K. I. (1981). Criterion bias and search sequence bias in word recognition. Memory & Cognition, 9(1), 78–92. https://doi.org/10.3758/BF03196953
Politzer-Ahles, S., Lee, K. K., & Shen, L. (2020). Ganong effects for frequency may not be robust. The Journal of the Acoustical Society of America, 147(1), EL37–EL42. https://doi.org/10.1121/10.0000562
Rayner, K., Ashby, J., Pollatsek, A., & Reichle, E. D. (2004). The effects of frequency and predictability on eye fixations in reading: Implications for the EZ Reader model. Journal of Experimental Psychology: Human Perception and Performance, 30(4), 720.
Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 81(2), 275–280. https://doi.org/10.1037/h0027768
Salamoura, A., & Williams, J. N. (2007a). Processing verb argument structure across languages: Evidence for shared representations in the bilingual lexicon. Applied Psycholinguistics, 28(4), 627–660. https://doi.org/10.1017/S0142716407070348
Salamoura, A., & Williams, J. N. (2007b). The representation of grammatical gender in the bilingual lexicon: Evidence from Greek and German. Bilingualism: Language and Cognition, 10(3), 257–275. https://doi.org/10.1017/S1366728907003069
Simpson, G. B., & Krueger, M. A. (1991). Selective access of homograph meanings in sentence context. Journal of Memory and Language, 30(6), 627–643. https://doi.org/10.1016/0749-596X(91)90029-J
Summerfield, C., & Egner, T. (2009). Expectation (and attention) in visual cognition. Trends in Cognitive Sciences, 13(9), 403–409. https://doi.org/10.1016/j.tics.2009.06.003
Thompson, M. C., & Massaro, D. W. (1973). Visual information and redundancy in reading. Journal of Experimental Psychology, 98(1), 49–54. https://doi.org/10.1037/h0034308