Nowadays, mental health is an increasingly important issue and thus drawing more and more attention from researchers in the field of linguistics (e.g., Hunt and Brookes Reference Hunt and Brookes2020; Price Reference Price2022). Collins and Baker's latest publication Language, Discourse and Anxiety broadens the scope of research on mental health as it conducts a corpus-assisted discourse analysis of conceptualisations of anxiety using a large-scale dataset collected from an online forum over a nine-year period. The word anxiety is the top keyword when the corpus is compared against the English Web 2020 corpus (a 38-billion-word general corpus), and the book identifies discursive patterns in forum members’ characterisation of anxiety and their cultivation of community of peers. It consists of six chapters besides introductive and conclusive chapters.

In the Introduction, the authors first identify the key types, symptoms, and treatments of anxiety and review previous studies on language and mental health, with a focus on those combining corpus linguistics and discourse analysis. Second, they explain the methodology used in this book, namely Corpus-Assisted Discourse Studies, after describing corpus techniques including frequency, keyness, collocation and concordance. Finally, the authors describe the data collection, namely the building of the corpus of about 21 million words with 294,082 comments collected from the Anxiety Support Forum between 2012 and 2020. The Introduction ends with an overview of the remaining chapters of the book to follow.

Chapter 2 explores the overall representation of anxiety. The authors conduct the collocation analysis of anxiety supported by the Word Sketch function in Sketch Engine, which organises collocates of the search term according to grammatical relations. Based on the automatic organisation and manual categorisation of collocates, four oppositional representation pairs around anxiety are identified, including medicalising versus normalising, catastrophising versus minimising, anthropomorphising versus abstracting, and owning versus distancing. Specifically, medicalisation is realized by patterns such as have anxiety and anxiety disorder while normalisation involves rare representations around what anxiety is or is not, such as ‘anxiety is normal’ and ‘anxiety is not an illness’. As for the second pair – catastrophising versus minimising, advice-seeking posters catastrophise anxiety mainly through adjectives describing its strength and length such as severe and constant, whereas those giving advice tend to minimise its effects through adjectives characterising its amount and popularity such as little and common. Similarly, advice-giving posters tend to anthropomorphise anxiety as a living agent through verbs including [anxiety] causes and [anxiety] loves, while anxiety is abstracted as a negative entity through abstract nouns as in [anxiety] disorder and [anxiety] cycle. As for the last pair – owning versus distancing, owning is typically realised by possessive personal pronouns, my/ our/ your anxiety, whereas distancing involves using the definite article the [anxiety] to increase a sense of detachment.

Chapter 3 explores the representation of the lived experience of anxiety by examining similarities and differences between Main posts and Replies. First, Main posts and Replies are separated and then compared against the English Web 2020 Corpus to derive the top 100 keywords and key terms, which are later classified into 10 thematic categories. The analysis of the shared key items in each category indicates that all posters tend to highlight anxiety's legitimacy and normalcy. The examination of the distinctive ones reveals that those in Main posts tend to emphasise the negativity of anxiety while authors of Replies are more positive in offering guidance. Finally, the authors explore key 6-grams and 11 frequent emoticons and emojis in each sub-corpus and find that creators of Main posts tend to emphasise the severity of anxiety with the use of negative emotion characters, whereas those in Replies highlight affiliation by using positive ones.

Chapter 4 begins by comparing Main posts with Replies against those without Replies to generate their unique keywords and examine their differences. The keyword analysis highlights that Main posts with Replies feature more reply-encouraging terms describing severity and urgency. Next, to track the articulation of the communicative purposes of forum posts, Egbert et al.'s (Reference Egbert, Wizner, Keller, Biber, McEnery and Baker2021) coding framework for discourse units (DUs), originally developed for use with spoken data, is adapted to online forum data. Each DU, with an identifiable beginning and ending, has one or more communicative purposes. Nine types of communicative purposes are coded on a scale from 1 to 3, with 3 standing for the dominant purpose in the DU and 1 or 2 representing less prominent purposes. The coding of 822 sample posts identifies 1,316 DUs which are further explored in terms of DU patterns, combinations of DUs within posts, and sequences of functional DUs. The exploration of DU patterns finds that the three most common communicative purpose are sharing feelings and opinion (fel), describing or explaining: time neutral (des), and giving advice and instructions (adv). Next, the analysis of combinations of DUs within posts reveals that the combinations of dominant communicative purposes fel+adv/des are frequent in posts with two or three DUs to reiterate the prevalence of advice-giving (adv) supported by personal experience (fel) and explanations (des). Finally, in exploring sequences of functional DUs in a discussion thread, the authors code 103 Main posts with 158 DUs and group them into five categories based on the prevailing communicative purposes across the DUs in the post. The analysis shows that the top three communicative purposes in direct replies to each category are fel, des, and adv, suggesting the forum's communicative objective as a resource to find advice and solidarity in struggling with anxiety.

To examine gendered differences in the forum, a direct comparison of male and female posts is first conducted in Chapter 5 to extract the unique keywords for each gender. The analysis reveals that men tend to view anxiety as a distant temporary state and focus on its impact on themselves, whereas women in general regard anxiety as a personal quality and consider it in societal structures. Second, a concordance analysis of gender-related words in the whole corpus indicates that women are represented as more likely to experience anxiety while men are viewed as less emotionally expressive and reluctant to acknowledge anxiety problems. Chapter 6 investigates cultural differences between the UK and the USA by comparing two sets of posts directly against each other to obtain unique keywords for each country. Besides, concordance analysis of words directly referring to nationality is carried out to supplement the keyword analysis. It is found that Americans provide more details in describing symptoms compared to Britons and ascribe anxiety mostly to food availability and political climate. In contrast, Britons tend to minimalise their anxiety experience and blame inclement weather for anxiety.

While the analysis in Chapters 2–6 is carried out synchronically, that in Chapter 7 is conducted diachronically. Specifically, the last analytic section explores the effects of change over time on language use in the forum through three-fold analyses. First, nine lists of top 50 annual keywords are identified by comparing each year against the other sub-corpora combined to examine the influence of external events on people's experience of anxiety. The analysis shows that only the COVID-19 pandemic in 2020 has a significant impact on the forum discussion. Then, the top 200 lexical words across the whole corpus are identified and grouped into 12 thematic categories to examine year-on-year change. The distribution of the top 200 lexical words in the annual sub-corpora reflects that the most common categories Body, Symptom, Medication, Test and Doctor have increased over time due to increasing numbers of American posters who focus on medications and symptoms. Second, the top 100 keywords for each age group are obtained by comparing each group against the other sub-corpora combined to examine age-graded differences. The findings reveal that the 20s group uses more hyperbolic and negative language to seek advice whereas the older groups employ more moderated language to give advice. Third, the keywords associated with different stages of posting are identified by comparing each stage's posts against the rest of the corpus to examine the language of posters across their journey from their first post to their last. The analysis indicates a progression from a troubled first post to longer-term contributors assuming supportive roles.

Chapter 8 concludes with reflections on methodological limitations, major findings regarding conceptualisations of anxiety, management strategies, and the value of the forum for posters, as well as suggestions for further research.

Overall, the book is an important contribution towards the fields of corpus linguistics, discourse studies and health communication. Five features of this volume stand out. Methodologically, except Chapter 2, the analysis is largely corpus-driven based on the keyness technique, starting not with predetermined research questions, but with questions emerging gradually in the analytic process. Besides, key multi-word terms such as panic attack and mental illness are extracted in most keyness analyses to identify concrete concepts and expressions that are more difficult to discern based on individual keywords. Furthermore, the authors not only rely on statistical measure to determine keyness, but also consider each keyword's textual diversity (range) to ensure its representativeness of the corpus. Structurally, each analytic chapter assumes the structure of an independent article, allowing readers to begin with any chapter in which they are interested. Finally, the authors make the analysis transparent and reliable by describing the analytic process as honestly as possible. For instance, the authors not only admit that the categorisation of keywords is subjective, but also report rare exceptions in concordance analysis whereas the corpus-assisted studies generally stress the overall trends. A minor limitation is that the unprecedented application of Egbert et al.'s (Reference Egbert, Wizner, Keller, Biber, McEnery and Baker2021) coding scheme in online forum data could make it relatively difficult for readers to follow, although the employment of the scheme sheds light on the creation of a community in an online forum.

In sum, the book fulfils its research aims through innovative and in-depth analyses, making it a unique and original contribution towards the characterisation of anxiety in online forums. Therefore, it not only provides a model for postgraduates and researchers to undertake studies of similar materials in the fields of corpus linguistics and discourse studies, but it also includes feasible suggestions for therapists, doctors, and lay readers.