Book announced at https://linguistlist.org/issues/36-1814

Title: Psycholinguistics
Subtitle: A Very Short Introduction
Series Title: Very Short Introductions
Publication Year: 2025

Publisher: Oxford University Press
http://www.oup.com/us
Book URL: https://global.oup.com/academic/product/psycholinguistics-9780192886774?utm_source=linguistlist&utm_medium=listserv&utm_campaign=linguistics

Author(s): Fernanda Ferreira

Reviewer: Victoria Beatrix Fendel

Fernanda Ferreira (2025). Psycholinguistics: A very short introduction. Oxford: Oxford University Press. ISBN 9780192886774. DOI https://doi.org/10.1093/actrade/9780192886774.001.0001.

SUMMARY

Ferreira’s Psycholinguistics: A very short introduction consists of nine chapters, of which the final one is an outlook chapter, followed by a subject index and a list of references and further reading. The book is intended to be an accessible and short, yet insightful and critical, introduction to psycholinguistics as a discipline. This goal is well achieved.

Chapter 1 introduces the reader to psycholinguistics as the study of “how humans process language and how they store and retrieve linguistic information” (p. 1) and subsequently provides a brief history of the subject and its influential theories and players. The chapter starts from the famous sentence the horse raced past the barn fell which will result in those unfamiliar with it getting garden-pathed (pp. 11–12), i.e. needing to revisit raced after reading the sentence and to reanalyse it as a participle. This shows how psycholinguistics connects linguistic concepts (e.g. participle) with psychological concepts (e.g. event) and cognitive processes (e.g. memory and decision-making). The dominant paradigm in linguistics prior to the 1950s was descriptivism and the dominant paradigm in psychology (esp. in North America) behaviourism (e.g. Skinner (1957)). Chomsky’s 1965 Aspects of the Theory of Syntax instead put forward a mentalistic and computational approach to language, which involved transformational rules applied to a base sentence such that the Derivational Theory of Complexity (DTC) was born (and subsequently disproved). Chomsky’s 1981 Lectures on Government and Binding “posited that complex sentences were derived via movement of a constituent from one location to another” (p. 9) and the moved constituent would leave behind a gap (this was experimentally tested with so-called Event-Related Potentials). Subsequently, lexicalist approaches to syntax put more emphasis on rules stored with lexical items (esp. verbs) as syntactic requirements vary by verb. In the 1970s, the Garden-Path model of language processing considered how ambiguity is processed, i.e. serial activation and subsequent repair if necessary (previously preferred) or parallel activation and subsequent discarding of the irrelevant meaning. The latter option was eventually favoured. Parallel models were more lexicalist in that they assumed that items were stored in the lexicon with their meaning and syntactic preferences. For instance, the alternatives for believe “have different strengths, proportional to their frequency of use” (p. 12). In the 1990s, new approaches hit the ground. Goldberg’s Construction Grammar appeared, Kirby’s work on artificial languages suggested that “grammatical rules have an abstract form because this facilitates communication” (p. 13), and computational linguists began work on machines that understand and produce language based on statistical patterns. Psycholinguistics moved away from the strict focus on syntax and now engages e.g. with prosody and pragmatics.

Chapter 2 is about processing (understanding) language and the mechanisms involved, when we speak and read about 240 words per minute. The process of word recognition is incremental. The Cohort Model (1980s) suggests that “the language system activates all words compatible with the current input until the word’s uniqueness point is reached” (p. 17). Context can shift this uniqueness point e.g. due to co-articulation or differing pronunciations based on accent, etc. The McGurk Effect refers to the fact that visual and acoustic cues are combined in language processing. We draw on contextual cues in order to deal with semantically ambiguous items. Multimorphemic items seem to be decomposed (e.g. rehouse). The parser, “the part of the language processing system that builds syntactic structure” (p. 20) is highly efficient with regard to repairs of garden-path sentences. When testing language processing experimentally, one may draw on the offline measure of acceptability, i.e. the participant will comment on the product of processing, or the online measure of tracking eye movements as participants read, i.e. the participant is monitored during processing. The language processing system draws on prediction, i.e. using past events in order to anticipate what is coming. It remains to be shown “how the processing of past versus upcoming linguistic content is coordinated” (p. 24). Early models of language processing are the Garden-Path model of human parsing, assuming that the processing module has no access to information other than grammatical rules, and lexicalist models, which assume that grammatical information is stored with word meaning. All these assume a serial parser. By contrast, Parallel Constraint-Based models assume that multiple interpretations are built in parallel and the strength of each option depends on its frequency. The Good-Enough language processing model considers, in addition to the compositional interpretation of a sentence “comprehenders’ expectations as a source of information that can generate interpretations” (p. 28). Noisy Channel models of language processing assume that the “language comprehension system has adapted to the reality that our cognitive and linguistic systems are not perfect”, i.e. there is noise such as distortions, deletions, or additions in the channel (p. 28). Surprisal, i.e. the expectedness of a word, correlates with “activation in cortical regions of the brain associated with language processing” (p. 30); entropy refers to “the uncertainty of a particular outcome given the current state” (p. 30). We can link both measures to processing difficulty.

Chapter 3 is about producing language and the mechanisms involved, including the retrieval of the word, choice of the grammatical structure, and pronunciation. Speech errors reveal that we plan ahead, hence why sentences like I got the park trucked appear at times. According to the two-stage model of language production, a speaker selects the concepts/lemmas and the syntactic skeleton and subsequently generates the sound pattern (word forms). Furthermore, an utterance may receive a specific prosody (e.g. the nuclear stress rule tells us that the main stress falls onto the final constituent in an English declarative). Thus, language production seems to be based on a slots-and-fillers architecture (p. 35). Experiments include e.g. Bock (1986), who showed participants an image, i.e. lightening striking a church, and a printed word phonologically or semantically related to church or lightening (a prime) in order to elicit active or passive sentences. Only the semantic primes had an influence. (Yet, a later experiment revealed that word beginnings were more influential than word endings and that the thus phonologically primed item tended to appear late in the sentence (p. 43).) The Bock experiment lends support to the two-stage model but also to the Easy-First model, i.e. “the production system is biased to begin a sentence with a word or concept that is accessible or easy to encode and produce” (p. 38). Easy-First allows the speaker to fluently go about their utterances rather than insert pauses and fillers (to gain processing time). Similarly, speakers insert that in complement clauses “when the next word or phrase is difficult for them to retrieve (p. 39). Participants can also be primed structurally whereby conceptual overlap enhances the effect (p. 41). There is some evidence for an internal editor that seems to verify whether what we are about to say is appropriate and which makes edits if it is not. This is an area of debate, and some have suggested that it is “the comprehension system applied to the output of our own production systems” (p. 44).

Chapter 4 combines language processing and production in the context of conversation and dialogue, our everyday real-life application of language. Grice’s maxims of truthfulness, relevancy, briefness, and clarity are comprehenders’ “starting point for their interpretations” (p. 46) in order to e.g. ascertain whether something is sarcasm. The maxims motivate the observation that “pragmatic inference is at the centre of communication” (p. 47) which Levinson argued promotes efficiency. Clark argued that a conversation is like “a duet” that speakers play on e.g. a piano (p. 48), it requires constant interaction and cooperation. Fillers for instance indicate planning problems to the interlocutor which may be related to something that is less familiar being about to be conveyed. Determining unfamiliarity depends on common ground between interlocutors, something we actually constantly need to test in teaching (p. 50). The common-ground theory also motivates the given-new strategy of ordering pieces of information. Audience design is the idea that language users “tailor their language to be as understandable as possible” (p. 51). Yet language users do not seem to do so as e.g. experiments relating to that inclusion in English show. This seems to be dependent on Proactive Interference (PI), “the negative effect of semantic similarity on the likelihood of retrieving information” (p. 53), for the speaker. Thus, it seems that audience design is a skill that needs to be acquired rather than being innate. Communication is successful when interlocutors are aligned (interactive model), i.e. have the same understanding of some topic of phenomenon (p. 54). This ties in with “prediction during language comprehension and self-monitoring of our own speech” (p. 55). It also ties in with entrainment, i.e. the tendency to model our linguistic behaviour on that of our interlocutor. This is largely automatic and unconscious. The Production-Distribution-Comprehension (PDC) model suggests that language users choose linguistic forms to maximise fluency. This ties in with the Easy-First strategy. The model, unlike the interactive model, also underlines that we learn based on frequency from input throughout our lives.

Chapter 5 is about reading which involves the decoding of “otherwise meaningless visual symbols” (p. 59) for sound and meaning. We learn how to read through instruction. Neural plasticity allows our neural systems to adapt. The Visual Word Form Area is situated in the left occipito-temporal cortex of the brain. The movements of the eyes during reading are called saccades (ca. 20 milliseconds). They happen between the fixations when the eyes stop moving (ca. 250 milliseconds). Eye movement can be, and often is, regressive when necessary. We can “extract information from approximately 3–4 letters to the left of the saccade landing site and 12–16 characters to the right” (p. 62) in English. This span can be divided into the foveal area (6–8 characters) where we get good visual detail, the parafoveal area (next 3–4 characters to the right) where we process characters for general shape, and the periphery (the remainder) which helps us find the point for the next fixation (p. 62). The perceptual span is noticeably asymmetric (across writing systems) and “reflects a combination of sensory, attentional, and processing factors” (p. 63). While for English reading involves an orthography-phonology-meaning route (except for common items), for a system like Chinese there seems to be an orthography-meaning route and phonology is activated optionally (p. 66). The perceptual span in English is larger than in Chinese possibly due to the information density as conveyed by the writing system (p. 67). The chapter closes with some comments on speedreading techniques, focussing on items in the middle of a line and Spritz, a technique that involves showing one item at a time. The latter reduces eye movements but does not allow for regressive movements or to build a non-flat prosody of the sentence. The former does not allow us to read for detail but results in significant guesswork along the way.

Chapter 6 is about individual differences in language processing. This has implications for learning, instruction, and remediation, for instance (p. 74). Abilities can be divided into those that are domain-general, i.e. any cognitive system can use them (e.g. speed of processing), and those that are domain-specific, i.e. only the language system can use them (e.g. vocabulary database). The size of the working memory (for many of us seven items (p. 71)) differs between people and is a domain-general ability but there is also the domain-specific verbal working memory (p. 76). One task to test working memory is Daneman and Carpenter’s task that “required subjects to hold and simultaneously process information” (reading span task) (p. 77). There are some issues with how the reading span task was administered because “working memory span is a continuous variable” rather than a categorial one and needs to be treated as such (p. 79). Furthermore, it is not clear whether the task measures working memory capacity or language processing skill (p. 79). Two electrophysiological events, so called event-related potentials (ERPs), benefit from an individual differences approach – the N400 (a negative deflection in the EEG signal, typically observed in relation to incongruities) and the P600 (a positive deflection in the EEG signal, typically observed in relation to repair events) are of particular interest (p. 81). Kim et al. (2018) “speculated that those with higher verbal working memory spans would exhibit the P600 response and those with lower verbal working memory spans would show the N400 response” (p. 82). Conversely, assuming that “the reading span task is a language processing task”, the argument has been made that “working memory is not real” but that it is all down to experience, yet the issue of cognitive ageing complicates this picture (pp. 83–84).

Chapter 7 is about the bilingual mind that deals with two language systems. This is “not exotic or unusual” (p. 85). Bilinguals can have a range of profiles with different levels of proficiency in each language they regularly use, and languages can be assigned to specific contexts of use. Based on experiments, it seems that word representations in response to for instance an image get co-activated in both languages (p. 87). There is the cognate advantage effect which suggests that finding the item in language A is easier when the item in language B is the same in sound, spelling, and meaning (p. 88). Work using the Visual World Paradigm has shown that the “less dominant language pokes through somewhat less” (p. 88). Code-switching refers to the scenario when an item, phrase, or the rest of a sentence are inserted in a language other than the sentence language preceding the switch. Code switches seem to be “perceived neurally as less weird or disruptive when the bilingual interacts with other bilingual speakers” (p. 90). Blending in code-switching, i.e. for instance repeating what was said already in the language switched into in order to make the syntax work, is common (p. 91). A key debate revolves around whether bilingualism has cognitive advantages, a position favoured in recent discourse, or disadvantages, a position previously favoured and rooted in nationalist, racist, and anti-immigrant discourses (p. 94). Questions of whether code-switching is similar to task switching in other cognitive domains have been asked along with questions about the ability to monitor and manage information, and inhibition – all these are subsumed under executive functions (p. 92). Bilinguals seem to show smaller effects, as compared to monolinguals, when it comes to dealing with incongruity – e.g. the Stroop effect (think of a printing the word “green” in red ink and asking what colour the ink is) or the Simon effect (think of a stimulus on the left-hand side of the screen and requiring the right hand to indicate detection) – and task-switching (e.g. a card-sorting task). Positive effects of bilingualism on creativity and metalinguistic awareness have also been reported (p. 93). However, the link between a cognitive reserve, a term “coined in the 1980s to describe older adults who at the time of their death showed few outwards signs of cognitive decline despite neurological evidence of Alzheimer’s disease” (p. 93), and bilingualism has received most attention. The idea is that challenging the brain (and body) will aid the creation of cognitive reserve (p. 94). However, the question is whether cognitive mechanisms involved in e.g. code-switching are domain-general and are thus also involved in e.g. task-switching from driving to talking (p. 95). By now, many studies report “no difference between monolinguals and bilinguals in executive functioning tasks” (p. 96). The previous non-availability of such results may have to do with the publishing industry’s aversion to publishing null results (p. 96). Furthermore, participants for studies “arrive at the lab already bilingual or not”, i.e. they are not randomly assigned to those conditions (p. 97), and bilingualism and immigration status are often confounded (p. 97). A general study design issue is the focus on undergraduates, due to their availability to the researcher (p. 98) – yet they are rather young and executive control is thought to peak around age 25 – and the lack of acknowledgement that there are many different kinds of bilinguals that are not necessarily comparable with each other. Yet even if we cannot prove a cognitive advantage, bilingualism certainly brings social and societal advantages (p. 99).

Chapter 8 is about manual-visual modalities of language, including sign languages and gestures. There are about 150 different sign languages. In these, signs represent meaning by “using arbitrary symbols” (p. 100) and possibly gestures alongside. Signs are defined with regard to “the shape of the hand, the location of the sign relative to the body, the movement of the hand through space, and hand orientation” (p. 101). The language functions and memory encoding of sign language seems to be the same as for spoken language (p. 101). Yet, processing of a sign is “less sequential than is the processing of spoken words” (p. 102). Unexpected signs, like unexpected words, “trigger an N400 response” (p. 102). Pauses indicate sentence, phrase, and word boundaries (p. 103). Semantically important distinctions, such as that between reciprocal and symmetrical events (e.g. kiss), can be encoded syntactically (p. 105). There seems to be no difference between signed and spoken languages as regards working memory tasks, once experimental set-up issues are resolved (p. 106). The ability to perceive visual information in the periphery seems to be more developed in deaf than in non-deaf individuals possibly due to “neural reorganization” (p. 107). Bilingualism between spoken and sign languages is bi-modal and code-blends, i.e. generating a spoken word and a sign simultaneously, are common. This seems to be a case of “co-activation without competition” (p. 109). We still know relatively little about how (often iconic) gestures that accompany a sign are produced and interpreted (p. 110). Özyürek’s Interface Hypothesis “assumes that language users generate their intention to produce speech and gesture at the same time” and gestures appear to mirror language-specific event structure encoding (pp. 110–111). Gestures are an integral part of (multi-modal) communication e.g. making the interlocutor aware that it is time to listen or facilitating the communication of complicated concepts (p. 112). Incongruent gestures trigger an N400 response (p. 111).

Chapter 9 is an outlook chapter primarily concerned with Large Language Models (LLMs) and Information theory. LLMs, e.g. ChatGPT, “interpret and generate human-like text” (p. 113). These so-called neural networks are trained on large amounts of data and “learn by identifying patterns” (p. 114). They can generalize from the learning context to other contexts. In the context of exposure to linguistic material having an effect on our processing of information, it is noticeable that LLMs function by “assigning probabilities to upcoming input based on previous exposure” (p. 114), yet they are trained with additional feedback techniques. LLMs do not have access to linguistic rules or principles (p. 115). Noticeably, they struggle with processing garden-path sentences, for instance (p. 115), and are trained on more data than a human will ever encounter in a lifetime (p. 114). LLMs produce at times “bizarre and even disturbing content” (hallucinations) suggesting that there is quite some mismatch between “how they work and how human process language” (p. 115). Finally, there are significant ethical challenges to do e.g. with reinforcing bias present in the training data (p. 115). Information theory emerged in the context of developing telephone networks (1940s) which posed the constraint of the physical structure to convey the message to be communicated. The focus is on efficient communication, that is “the message received corresponds closely to the message sent, and has been derived with as little effort as possible” (p. 116). This brought into focus the concept of the noisy channel too, i.e. interlocutors make mistakes or environments are literally noisy. Faced with this noisy channel the processing system will “use its expectations” (p. 117) to arrive at a decoding of the message. As regards brain mechanisms and language processing, we are nowadays interested in knowing where in the brain language functions take place (and where it does not) (p. 118). An interesting finding by Fedorenko and Kanwisher (2009) in this context is that “language and thinking are not the same thing, which means that it is indeed possible to have complex thought without any language abilities” (p. 119). Finally, psycholinguistics is moving away from the focus on the WEIRD (Western, Educated, Industrialised, Rich, Democratic) participant and embracing diversity (p. 119). The logistics of non-WEIRD speaker communities however will require adaptation of our established processes and techniques (p. 120). Participants should never be just sources of data but partners in “a respectful, mutually beneficial exchange” (p. 121).

EVALUATION

The book reads effortlessly, without any garden-path sentences, throughout despite the density of information content, as becomes especially evident when one summarises chapters after reading them. It presents concepts, frameworks, and theories from multiple perspectives, and explains how experimental designs and results, which are vital in psycholinguistics, were and are translated into further reaching generalisations. The book succeeds in introducing the reader not only to theories, concepts, and frameworks, but also to the big names in the field. The subject index makes it possible to use it as a quick-reference guide in pocket format (the print version is postcard-sized).

The book taps into several topics of psycholinguistics that have filtered down into general public discourse. The first is speedreading techniques. Ferreira debunks why these techniques will never result in full understanding – due to guesswork, due to avoiding regressive eye movements, and due to suppressing a non-flat prosody (see Chapter 5). The second is the famous cognitive reserve, which is why people do crosswords and learn languages on Duolingo for instance. Ferreira highlights the social advantages of bilingualism and language learning – including breaking down communication barriers and promoting social harmony (p. 87) – over the cognitive ones and thus underscores the importance of bilingualism as a societal tool (see Chapter 7). The third is the emphasis on WEIRD subjects in participant-based studies. These studies can be, and often are, perceived as test environments (see e.g. Vieregge (2025: 238) for a participant’s comment to this effect) which makes people weary of signing up, yet in all reality researchers are fundamentally interested in diversity and individuality, and changing the discourse surrounding such studies could change the rate at which participants come forward to get involved (see Chapter 9). Finally, a more research and publishing industry inherent observation is made with regard to null results and the likelihood, or lack thereof, of these getting published and received. This observation is pertinent if we want to avoid bias in the research record that the field is building on in future, and more awareness of the issue may well move the goalposts.

I have three minor comments on content and form that I briefly set out below. My first relates to Chapter 7 on bilingualism and bilinguality. Ferreira (pp. 85–86) sets out clearly that bi- and multilingualism is not exotic or rare and that bi- or multilinguals have a wide range of proficiency profiles. Not all of them began learning two languages the moment their lungs first filled with air nor do all of them have a migration background, to name only two of the most common misconceptions. One may add Grosjean’s language modes to these characteristics of bi- and multilinguals, especially given the observation that code switches seem to be “perceived neurally as less weird or disruptive when the bilingual interacts with other bilingual speakers” (p. 90). At the edges of the language mode continuum, there are the monolingual and the bilingual modes. In the monolingual mode, the language that is not used in active production is deactivated; in the bilingual mode, both languages are activated (Grosjean 2024: 28). A bilingual will shift easily towards the bilingual end of the mode continuum when they know that e.g. the interviewer is bilingual or when they notice the setting being favourable towards bilingualism. It is thus a variable to control for in experimental studies (Grosjean 2024: 31 and 48). Thus, in addition to the varying proficiency profiles, the surroundings of the experimental settings are likely to have had an influence along this mode continuum in the past without this being controlled for. This makes results at least difficult to compare.

My second minor comment relates to Chapter 9 on computational approaches to linguistics and specifically neural networks and Large Language Models. Ferreira (pp. pp. 114–115) sets out clearly the differences between a human processor and an LLM: LLMs see vastly more data than a human being in a lifetime, LLMs cannot process garden-path sentences for instance (perhaps due to the lack of access to rules?), and LLMs hallucinate, i.e. create bizarre or disturbing content. Especially the emphasis on the quantity of training data and the resulting output is something that has not filtered down into general discourse. One could add to this list also that LLMs cannot draw inferences based on the social, political, cultural, or personal surroundings or localise beyond something akin to autumn as opposed to fall or behaviour as opposed to behavior. It almost seems like there is a link between the linguistic sign and the thing it designates, but there is no link between these and the user/interpreter (cf. Culpeper 2021: 17). This absent link with the user may also make it impossible for an LLM to access the non-referential function of the linguistic sign, so-called indexicalities referring to societal discourses (cf. Eckert 2008). This is especially pertinent as the training data that LLMs are drawing on contains significant biases that are reinforced through training LLMs on this data (pp. 115–116). Policymakers are still lagging behind the technological development in order to ensure ethical usage (cf. e.g. Gajjar & Brione 2024; Kashefi, Kashefi & Ghafouri Mirsaraei 2024).

Finally, the only sentence in the entire book that required a double take is the following: (p. 120) “the work that scientists like Majid is doing are a critical step in that direction.” The verbs “is” (subject “the work”) and “are” (subject “scientists like Majid”) are swapped incorrectly. Given the multiple production stages involved beyond the writing and editing processes, thinking about long-distance dependencies or planning ahead while producing language may not be at play here, yet the reader actually thinks back to these concepts when encountering this sentence. Thus, in the end, this may well be a clever twist to make the reader think.

REFERENCES

Bock, Kathryn. 1986. Meaning, Sound, and Syntax: Lexical Priming in Sentence Production. Journal of Experimental Psychology: Learning, Memory, and Cognition 12(4). 575–586.

Culpeper, Jonathan. 2021. Sociopragmatics: Roots and Definition. In Dániel Z. Kádár, Marina Terkourafi & Michael Haugh (eds.), The Cambridge Handbook of Sociopragmatics, 15–29. Cambridge: Cambridge University Press.

Eckert, Penelope. 2008. Variation and the indexical field. Journal of Sociolinguistics 12(4). 453–476.

Fedorenko, Evelina & Nancy Kanwisher. 2009. Neuroimaging of Language: Why Hasn’t a Clearer Picture Emerged? Language and Linguistics Compass 3(4). 839–865.

Gajjar, Devyani & Patrick Brione. 2024. Artificial intelligence: ethics, governance and regulation. UK Parliament POST. https://post.parliament.uk/artificial-intelligence-ethics-governance-and-regulation/.

Grosjean, François. 2024. On Bilinguals and Bilingualism. Cambridge: Cambridge University Press.

Kashefi, Pouya, Yasaman Kashefi & AmirHossein Ghafouri Mirsaraei. 2024. Shaping the future of AI: balancing innovation and ethics in global regulation. Uniform Law Review 29(3). 524–548.

Kim, Albert, Leif Oines & Akira Miyake. 2018. Individual differences in verbal working memory underlie a tradeoff between semantic and structural processing difficulty during language comprehension: An ERP investigation. Journal of Experimental Psychology: Learning, Memory, and Cognition 44(3). 406–420.

Skinner, Burrhus. 1957. Verbal behavior. Illinois: Copley Publishing Group.

Vieregge, Annika. 2025. Bewertung und Variation der Präpositionalkasus im Deutschen. Berlin: Language Science Press.

ABOUT THE REVIEWER

Victoria B. Fendel (D.Phil. Oxford, 2018) is a research associate at the University of Oxford, one of the editors of the Classics section of the Literary Encyclopedia, and language leader for Ancient Greek in the PARSEME/UniDive COST initiative. Her research focusses on bilingualism and language contact (Oxford University Press, 2022), multi-word expressions (De Gruyter Brill, 2025) in literary, epigraphic, and papyrological sources, and on the development of digital tools for large corpora (Language Science Press, 2024).