The rapid expansion of digital corpora and computational methods has transformed the empirical scope of English linguistics. Against this background, Data-Intensive Investigations in English Linguistics, edited by Mikko Laitinen and Paula Rautionaho, brings together a range of case studies from an international group of scholars, exploring how data-intensive and computational methods can be used to address research questions grounded in the theoretical traditions of English linguistics. The book comprises ten chapters and covers a diverse range of topics, including phonetics, corpus construction, historical linguistics, word formation, syntactic variation and methodological reflection. While the chapters vary in their specific aims, datasets and methods of analysis, they demonstrate how large corpora and advanced analytical techniques can yield new insights into linguistic research questions.
The introductory chapter by Mikko Laitinen, Paula Rautionaho and Irene Taipale (pp. 1–19) sets the stage by outlining the historical, conceptual and methodological foundations of data-intensive linguistics. The authors define data-intensive methods, note an ongoing shift towards algorithm and tool development in English linguistics, and discuss ‘digital fetishism’, or the possibility that a focus on tools and data informatics could imperil theoretical inquiry and endanger the epistemological foundations of the field, rather than complementing and extending it. This framing is effective in positioning the volume within broader ongoing debates about the role of computational approaches in the humanities and provides a clear rationale for the case studies that follow.
Chapter 2 (pp. 20–47), by William A. Kretzschmar Jr, Michael Olsen, Rachel Olsen and Katherine Ireland, provides a corpus-phonetic analysis of monophthongs extracted from the Digital Archive of Southern Speech (DASS), a selection of dialectological fieldworker interviews conducted from 1969 to 1983 in the context of data collection for the Linguistic Atlas of the Gulf States (LAGS). After manual transcription, alignment, vowel extraction and measurement with the Montreal Forced Aligner and FAVE (Rosenfelder et al. Reference Rosenfelder, Fruehwald, Evanini, Seyfarth, Gorman, Prichard and Yuan2015; McAuliffe et al. Reference McAuliffe, Socolof, Mihuc, Wagner and Sonderegger2017), the authors organize the F1/F2 formant space into a grid and count the number of vowel measurements in each cell. They observe that a small number of cells contain many measurements, while many cells contain relatively few, a pattern interpreted as an ‘A-curve’, and argued to capture phonetic variation more effectively than reliance on mean values. The chapter demonstrates how large-scale phonetic datasets can be derived from naturalistic recordings and highlights the value of legacy dialect materials.
At the same time, several methodological choices raise questions. The absence of measurements in the lower-right region of the F1/F2 space is interpreted as a property of the speakers’ vowel realizations; however, this distribution follows from the inherent relationship between F1 and F2 values, rather than from speaker-specific patterns. In addition, no vowel normalization is applied to account for systematic differences across speakers, and the rationale for discretizing continuous formant data into grid cells, thereby reducing precision, remains unclear. The proposed ‘A-curve’ pattern is presented as a general property of the data; however, it is not clear to what extent this pattern reflects underlying phonetic structure, as opposed to being an artefact of the discretization procedure. More generally, the analysis does not engage with methods that are now standard in corpus phonetics, such as mixed-effects modeling or quantitative measures of vowel overlap (e.g. Pillai or Bhattacharyya statistics), which could have provided a more direct means of addressing some of the questions posed (for instance, whether particular vowel categories occupy overlapping regions of the acoustic space or shift over time). While the chapter raises interesting questions and makes productive use of a large dataset, a closer engagement with established methodological practices would have strengthened the analysis and clarified its contribution.
In chapter 3 (pp. 48–74), Keiko Bridwell and Katherine Ireland describe in greater detail the procedures used to process the DASS material, which also forms the basis for the analysis presented in chapter 2. They outline the use of tools such as Corpus Workbench, the Stanford part-of-speech tagger and the Montreal Forced Aligner to create transcripts annotated at the word level with speaker metadata. In the second part of the chapter, the authors analyze the use of the epistemic verb reckon in the DASS data by means of linear regression, finding that it is associated with older speakers in Alabama from lower socio-economic backgrounds and with less formal education, according to the original LAGS/DASS metadata. The analysis supports the claim that the item is strongly associated with Southern American speech, to the extent that it is used in media to construct a stylistic stereotype of the variety. Overall, the care taken in the processing of the legacy DASS data and in the analysis of reckon lends credence to the authors’ claim that the DASS corpus represents a fruitful resource for corpus-based analysis of Southern American English in the late twentieth century.
The fourth chapter (pp. 75–103), by Sarah van Eyndhoven, Lisa Gotthard and Rosa Filgueira, considers the use of Scots from the seventeenth to nineteenth century in a corpus of chapbooks (‘chapmen’s books’, a type of cheaply printed pamphlet containing a range of popular material) made available in digitized form by the National Library of Scotland. The core of the chapter details the use of defoe, a GitHub repository comprising Apache Spark and Python-based corpus analysis scripts for working with XML and related formats, to compare the frequencies of Scots and Southern English wordforms. After organizing the corpus material by location, topic and year, the authors employ conditional inference trees to analyze the extent to which Scots persisted in these texts. Rather than disappearing, the use of Scots is shown to increase over the period under study, particularly in the nineteenth century, a trend the authors attribute to the ‘Vernacular Revival’ of Scots in popular literature, even as it recedes in more formal prose. The chapter thus challenges the widely held assumption that Scots disappears in this period and emphasizes the continuity of its use in this genre. Its contribution lies both in drawing attention to an understudied type of printed material and in providing a quantitative account of Scots features, as reflected in lexical and orthographic variation. At the same time, the methodological infrastructure on which the analysis relies appears somewhat dated. The defoe toolkit does not appear to be actively maintained, which may limit its applicability in current research environments. More broadly, it reflects the fact that much of the work in the volume was carried out several years prior to publication, which affects the present-day relevance of some of the tools and workflows employed.
In chapter 5 (pp. 104–36), Gerold Schneider investigates the detection of idiomatic expressions through a combination of collocation measures, distributional semantic similarity and statistical tests. Using data from the British National Corpus and the English Corpus-as-Web (EnCOW), lexical association measures (O/E and O2/E) are combined with embedding-based cosine similarity to quantify the strength of collocations and idioms; results are presented as tables of combinations with high derived scores. The chapter shows that a linear model incorporating semantic similarity (cosine distance) alongside collocation measures improves the detection of collocations and idioms. While the approach is methodologically comprehensive, the presentation is at times dense, and some aspects of the exposition may be difficult to follow for readers without a background in distributional semantics.
Chapter 6 (pp. 136–66), by Sabine Arndt-Lappe, Natalia Beliaeva and Audrey Martin, focuses on word formation, specifically the phenomenon of name blending. In a first study, using the large News on the Web (NOW) corpus, the authors calculate sentiment values for sentences in which seventeen ‘determinative blends’ occur – blends such as Borisconi, where Boris Johnson’s name is combined with Silvio Berlusconi’s, but the referent remains Boris Johnson. For this type of blend, they find that such forms tend to occur in contexts with negative sentiment. In a second study, a larger set of ‘coordinative blends’ is analyzed; these are blends used to denote couples or pairs (e.g. Brangelina). Coordinative blends are more likely to occur in contexts associated with positive sentiment. The chapter offers an interesting perspective on how quantitative approaches can be applied to areas traditionally associated with qualitative analysis. It also raises interesting questions about the relationship between productivity, frequency and acceptability in word formation, suggesting directions for further research.
Chapter 7 (pp. 167–95), by Paula Rautionaho and Lea Meriläinen, examines future time reference across native and non-native varieties of English. The authors analyze the relative frequencies of the forms will, ’ll, won’t, going to and gonna in the Spoken British National Corpus (Spoken BNC2014), the Corpus of Contemporary American English (COCA) and the Louvain International Database of Spoken English Interlanguage (LINDSEI), focusing in the latter on speakers from Czechia, Finland and Taiwan. They find higher use of going to in COCA, corroborating previous findings (e.g. Leech Reference Leech, Facchinetti, Palmer and Krug2003; Leech et al. Reference Leech, Hundt, Mair and Smith2009). Using a generalized linear mixed-effects model tree that incorporates intra-linguistic variables such as sentence type, clause type and polarity, the authors show, for example, that gonna is more likely to occur in main clauses than going to, and that the contracted form ’ll is more frequently associated with first- and second-person pronouns. Taiwanese speakers are more likely to favor will over going to compared to the other learner groups considered in the study. The chapter provides a clear and methodologically well-grounded analysis and demonstrates the value of large datasets for comparative studies of World Englishes, while also underscoring the importance of careful corpus design in ensuring comparability across datasets.
In chapter 8 (pp. 196–226), Juho Ruohonen and Juhani Rudanko apply Bayesian multivariate analysis to the study of complement selection, focusing on subject-control complements of the verb fear and considering the alternation infinitival versus gerundial complementation (e.g. fear to walk vs. fear walking). Starting from a grounding in theoretical syntax, the authors analyze 1,475 sentences from American English data from the NOW corpus with respect to fourteen features of the complement type. Noting the difficulties of using multilevel logistic regression on a dataset of this size if all variables are to be considered, the authors undertake a Bayesian analysis, setting model parameters in a way that allows similar parameters to a logistic regression model to emerge. The model results show that in general, the gerundial complement is favored. The form of the head verb (fear vs. fearing) has a strong effect: fear/fears favor gerundial complements, while fearing favors to-complements. The parameters for other covariates are interpreted in line with proposed principles of theoretical syntax. The use of Bayesian methods represents a relatively recent development in corpus linguistics, and the authors provide a clear rationale for their approach. The chapter is methodologically rigorous and contributes to ongoing discussions about the advantages of Bayesian over frequentist models. However, the technical complexity of the analysis, together with its strong grounding in generative syntactic theory, may limit its accessibility for many corpus-oriented readers.
Chapter 9 (pp. 227–57), by Maryam Nasseri, explores the statistical modeling of syntactic complexity in English academic texts. The study examines features such as subordination and clausal structure in a corpus of 210 master’s theses on linguistics written by L1, ESL and EFL students, analyzing how these features are distributed across rhetorical sections (e.g. Abstract, Introduction, Literature Review). The analysis employs four machine-learning methods to predict section type based on syntactic features. The findings show, for example, that coordination is more likely to occur in abstracts and in results and discussion sections than in other parts of a thesis. By identifying syntactic predictors associated with rhetorical structure, the study offers insights into the relationship between form and function in academic discourse. While the study identifies a number of distributional patterns, its broader contribution is somewhat limited by the methodological framework employed. The machine-learning approaches used reflect the state of the field at the time the research was conducted but now appear dated in light of more recent developments in computational linguistics, particularly the widespread adoption of LLM-based approaches. In addition, although the corpus includes L1, ESL and EFL writers, differences across these groups are not systematically explored.
The volume closes with Martin Schweinberger’s chapter (pp. 258–87), which addresses the implications of the replication crisis for data-intensive corpus linguistics. This contribution provides a valuable meta-methodological perspective, emphasizing the need for transparency, reproducibility and robust research design. The author advocates data transparency, for example through the use of platforms such as the Open Science Framework, GitHub and CLARIN, as well as the implementation of FAIR principles for data access. He further emphasizes the importance of consistent project design, the use of free and open-source software, and the availability of reports and code in reproducible formats (for instance, through R or Python notebooks). Many of these principles have been implemented at the Language Technology and Data Analysis Laboratory at the University of Queensland, Australia. A short analysis of amplifier alternation in Asian Englishes is provided to illustrate some of the desiderata outlined in the chapter. By situating corpus linguistics within broader discussions of scientific practice, the chapter highlights both the strengths and potential vulnerabilities of data-intensive approaches, serving as a fitting conclusion to the volume.
Overall, the volume offers a rich and varied set of contributions that demonstrate the potential of data-intensive approaches in English linguistics. Its principal strength lies in the breadth of its case studies, which span multiple subfields and methodological traditions, as well as in its consistent emphasis on empirical data and methodological transparency. The chapters are, for the most part, thoughtful and engaging, and together they provide a valuable overview of current directions in data-intensive research.
As is often the case with edited collections, the level of technical detail and accessibility varies, and the connections between chapters are not always made fully explicit. A small number of contributions also raise methodological questions; for example, chapter 2, which would have benefited from closer engagement with established practices in corpus phonetics, and chapters 4 and 9, which rely on tools and modeling approaches that now appear somewhat dated. These issues, however, are best understood in light of the volume’s origins in work presented at the 2021 ISLE conference. Given the rapid pace of development in computational linguistics, and particularly the subsequent shift toward newer modeling approaches, some aspects of the work inevitably reflect an earlier stage in the field’s development.
These reservations do not detract from the overall contribution of the volume. On the contrary, Data-Intensive Investigations in English Linguistics succeeds in showcasing a wide range of innovative and methodologically diverse approaches, while also engaging with important questions about the role of data-intensive methods in linguistics. The volume will be of considerable interest to researchers and advanced students in corpus linguistics, sociolinguistics and digital humanities, and it provides a useful foundation for further work in this rapidly evolving area. It stands as a timely and valuable contribution that will stimulate continued discussion and development in the field.
