The aim of the book reviewed here, Crossing Boundaries through Corpora: Innovative Corpus Approaches within and beyond Linguistics, is to bring together approaches to automating parts of or whole research workflows in a wide variety of linguistic subdisciplines – for a quantitative overview of these in the last two decades see Crosthwaite et al. (Reference Crosthwaite, Ningrum and Schweinberger2023). In doing so, this volume also successfully achieves what is frequently called for: integrating approaches and moving many of these subdisciplines closer together.
In the introductory chapter (pp. 1–6), the five editors describe their approach to collating the work of eighteen colleagues, aiming to push the boundaries of nine different research areas. They describe that this volume showcases automated and semi-automated approaches across phonetics, phonology, word formation, lexicon, syntax and semantic analysis as well as advances in corpus tool accessibility and corpus-based language teaching. The collection is organised around four types of ‘boundary crossings’: between scientific approaches, across temporal boundaries, through spatial and disciplinary boundaries via dataset repurposing, and by improving corpus accessibility through enhanced query interfaces.
The ICAME conference is the original venue which hosted the talks that serve as the basis for this edited volume and the first content chapter is positioned closely to the roots and spirit of this conference. The subarea of corpus linguistics that is in focus here is historical sociolinguistics, a field that has developed significantly since the emergence of diachronic corpus linguistics in the 1970s, and the present contribution by Tanja Säily, Martin Hilpert and Jukka Suomela constitutes a major development in the wider field of historical sociolinguistics (for previous work of a similar character see e.g. Whitt Reference Whitt2018). Titled ‘New approaches to investigating change in derivational productivity’ (pp. 8–40), this chapter offers a new methodological framework for visualising linguistic change while accounting for variations in corpus size. It is highly relevant since this is one of the field’s persistent challenges and demonstrates the usability of the approach in applying it to an exploration of diachronic changes in -ity versus -ness suffix use and productivity in British English between 1600 and 1800. Their findings confirm patterns observed across different corpora and registers in previous research, adding nuanced insights regarding possible root causes for gender differences in suffix usage by looking at type-based (etymological, word class and branching structure) as well as token-based (semantic type and occurrence in possessive constructions) variables. The authors have built a versatile research tool which enables the creation of graphs that communicate complex language changes effectively in meaningful, accessible formats, while ensuring statistical reliability through significance testing. This aspect, in combination with the commitment to open workflows and scalability, makes the methodology especially relevant not only for historical research but potentially valuable for contemporary low-resource linguistic contexts as well. Overall, this chapter presents a solid, well-constructed contribution to historical linguistics through its innovative methodological advancement of diachronic corpus visualisation techniques.
Chapter 3, by Martin Schweinberger and Yuki Komiya, then transports the reader to the present and into the domain of second language learning and teaching in ‘A corpus-based comparative acoustic analysis’ (pp. 41–61). More specifically, this chapter compares English vowel production in native English speakers to that of L1-Japanese learners using data from spoken monologues provided by a total of 255 participants in the International Corpus Network of Asian Learners of English (ICNALE). The innovative component of this chapter lies in the application of Gries & Dehors’ (Reference Gries and Deshors2014) Multifactorial Prediction and Deviation Analysis using Regression/Random Forests (MuPDARF) to the data and a thorough evaluation of the usability of the approach as well as an evaluation of factors contributing to non-target-like pronunciation. Schweinberger and Komiya show that difficulties for L1-Japanese learners of English are primarily confined to a small set of vowels (/ɪ/, /u/, /ʊ/ and /ɛ/), with problems increasing with longer vowel durations and in less frequent words; in contrast to this, /ɒ/, /æ/, /e/, /i/ and /ʌ / are produced with markedly lower error rates. The authors suggest that the higher error rates might stem from tongue raising issues that these learners might face ‘due to generally more centralized tongue positioning in Japanese’ (p. 55). While this chapter is clearly a good fit for the Crossing Boundaries through Corpora volume and can and should be used as a basis for future studies aiming to provide clear language teaching recommendations, the study also highlights wider audio quality issues in the field that, if addressed, would be a meaningful step towards more reliable cross-linguistic research.
Chapter 4 ‘Digital Dickens’ by Gerold Schneider (pp. 62–98) then takes the reader a step further, and into the realm of Digital Humanities in the shape of an in-depth discussion of Dickens’ novels in contrast to the work of his contemporaries, particularly focusing on how the author portrays poverty and social issues. A target corpus of eight novels by Dickens (1.6 million words) from the Gutenberg Archive is contrasted with the Corpus of Late Modern English Texts (CLMET 3.0) as the reference corpus. What makes this contribution stand out is the impressive range, evaluative depth and accessible description of a wide range of methodological approaches. A simple frequency approach of looking at poor and poverty in the selected novels is enriched by employing a successful Document Classification to predict if text snippets are part of the Dickens dataset or CLMET corpus (97.6% accuracy).
Moving towards a closer reading of the meaning behind individual items, the author employs a Distributional Semantics approach and extracts words that are semantically closest to poor in Dickens’ work and CLMET before moving on to Topic Modelling to identify themes in Dickens’ writing. The researcher’s commitment to methodological accessibility is evident in his clear explanation of complex techniques like Kernel Density Estimation (KDE) for creating ‘conceptual maps’ using the Python library textplot. This, already very dense, discussion of DH methods and their applications in the context of representations of poverty in Dickens is then further developed by exploring the results to identify whether Dickens’ style can be traced automatically and whether it shows the expected signs of literary realism. A special focus here is on conceptual maps, networks that represent associations between terms. The chapter’s computational approach to Dickens analysis complements established tools like CLiC (Corpus Linguistics in Context; Mahlberg et al. Reference Mahlberg, Stockwell, De Joode, Smith and O’Donnell2016), and champions true methodological triangulation and innovation which should inspire further work of a similar kind both in its openness to transgressing corpus linguistic borders and in integrating multiple computational approaches.
Reporting clauses, the clauses introducing quoted speech in a direct speech segment, are at the centre of Jarle Ebeling’s chapter in this volume, which can be categorised as a work in corpus stylistics (‘120 years of reporting clauses: Stability or change?’, pp. 100–24). In this chapter, the author introduces the newly compiled 120-million-word Corpus of British Fiction (CBF) as a resource to answer questions regarding changes in reporting verb usage between 1900 and 2019. The research employs a mixture of quantitative and qualitative methodologies to analyse patterns in direct speech reporting. A key contribution is Ebeling’s modified classification system for reporting verbs, categorising them into twelve distinct types including attitude, gesture, illocutionary, manner, mental, neutral, prosodic, prospective, retrospective, discourse-signalling and the verb say itself. The findings challenge initial hypotheses in showing that single-adverb expansions have decreased despite the rising dominance of say as a reporting verb. The overall picture emerging from the qualitative analysis is one of stability, rather than change, over the period of observation with regards to the use of ing-expansions. Qualitatively, in this corpus they predominantly encode body movements or mental activities which provide readers with additional contextual information about characters’ physical and psychological states during dialogue. The study addresses crucial methodological considerations, acknowledging the limitations of rule-based automated recognition of direct speech (with recall rates of approximately 70% against precision rates of 98%). Ultimately, Ebeling’s work illustrates the value in pushing disciplinary boundaries and integrating quantitative and qualitative approaches when investigating complex linguistic and stylistic phenomena in literary texts.
In the following chapter, ‘Establishing a “new normal”’ (pp. 125–52), Matt Gee, Andrew Kehoe and Antoinette Renouf propose a novel processing pipeline with the objective of creating accessible, statistically sound and visually appealing word frequency trend charts over time. Their approach refines earlier work and focuses more strongly on customisability, scalability and detection of lasting flexible timespan trends as opposed to registering statistical noise or temporary blips in usage patterns. The authors’ approach is demonstrated on a 1.8-billion-word British news corpus spanning over thirty years, from which they extract over 23,000 type candidates to consider. For selected types that show an interesting diachronic development, general frequency trend charts are created, and their collocational behaviour is visualised using horizon charts. What is particularly noteworthy is the combination of methodological innovation and, at times, a rather technical description of the processes at hand, with a good amount of concordance lines for the respective periods. This follows the general theme of the chapter in emphasising the value of collocational context for these analyses which, in turn, necessitates a thorough qualitative exploration alongside the quantitative component. The case studies demonstrate how certain lexical changes seem to reflect broader cultural shifts – for instance, the changing usage patterns of words like chairman in the context of the diminishing use of gendered language. Overall, this chapter provides valuable insights and starting points for researchers working with diachronic corpora, and the processing pipeline could readily be adapted to numerous research contexts beyond news media.
Chapter 7 takes the reader into the new approaches section of the book and changes course to look at a different domain: prosodic and syntactic analysis of spoken language (‘Syntactic segmentation of spoken corpus data: What prosody can contribute’, pp. 154–91). The authors, Karin McClellan, Kathrin Kircili and Sandra Götz, take a closer look at segmentation problems in spoken corpora and approach this topic from a new angle. Their fine-grained approach aims to improve segmentation by using prosodic information during transcription which they test on a subset of the Louvain Corpus of Native English Conversation. They outline that segmentation problems arise from the frequent occurrence of non-clausal units (such as ‘Why not?’) in naturally occurring speech. In order to remedy this, they suggest two classes of non-clausal units, inserts and syntactic non-clausal units, as well as their respective subcategories. These are then used in conjunction with Praat intonation contours for the sentences at hand to contrast the resulting prosodic information with a syntactic interpretation of the samples. They find that there is overall low overlap between prosodic and syntactic segmentation, with the least overlap for the highest level of granularity when transcribing. They further find that 9 per cent of the data required prosodic information for accurate analysis. While this is based on a very small sample of texts, this finding still indicates the need for larger-scale adoption of the syntactic cum prosodic analyses, especially in cases where the data at hand is insert-heavy.
In chapter 8, Tobias Bernaisch and Sven Leuckert leverage newspaper language to explore changes and constants in textual functionality found in South Asian English newspaper data (‘Short-term diachronic and variety-internal approaches to textual functionality in South Asian Englishes’, pp. 192–216). They achieve this by employing a Multidimensional Analysis (MDA, Biber Reference Biber1988) and undertaking a hierarchical cluster analysis on the resulting scores in order to explore a variety of World Englishes with a special focus on a nuanced discussion on the topic of plurality and uniformity in these varieties. Their research is based on three newspaper corpora: SAVE, SAVE2020 and CORINNE, where the former two datasets are taken to be representative of more strictly edited texts and therefore closer to standardisation, and the latter is taken to be less rigidly edited and therefore retaining more of the regional variation of language. Newspaper texts have specifically been selected here since they display a degree of codification that can be difficult to obtain from other sources for less formalised World Englishes. The authors introduce the methods in an accessible and comprehensive manner. What makes this contribution exceptional and innovative is the graded discussion of theoretical limits, such as labelling certain varieties as standards or quasi-standards semi-arbitrarily, which poses substantial problems for practical work in this area. Overall, the results show a slight trend towards using more speech-like and non-narrative features across the dataset, with the possible exceptions of Maldivian and Bangladeshi texts/samples, where the newspaper authors came to use more written features over time. A further core finding of the MDA cluster analysis is that the strongest effect on functional textual similarities is rooted in the regional provenance of a text, with comparably minor effects from language change over time. These findings provide crucial empirical evidence for debates about standardisation processes in World Englishes and the persistence of regional linguistic identities.
Maintaining the spotlight on world Englishes, Julia Schlüter presents a questionnaire study conducted with native English-speaking teachers for the ninth chapter, titled ‘Do corpus data on World Englishes inspire tolerance of variation in ELT professionals?’ (pp. 217–45). Her work effectively combines usage-based language learning and variations of World Englishes with a necessary and long overdue excursion beyond linguistics in a stricter sense: here, conducting a survey of 76 native English-speaking teachers who were prompted to give acceptability ratings of alternative linguistic features that are well established in other World Englishes but may differ from their native preference. The participants were all actively teaching English as a foreign language and were sampled via different German universities. This work falls into the same category as, for example, Curry & Mark’s (Reference Curry and Mark2024) study on necessary pedagogical considerations for corpus-based language learning practices and materials.
There are many arguments for using data-driven and corpus-assisted strategies in the classroom, but it is impossible to establish impactful practical tools and guidelines without considering where practitioners stand on these matters. Rejection of variationist perspectives, especially by conservative decision-makers in educational settings, is a known hurdle alongside issues surrounding the workload implications of accepting graded or alternative models of correctness. In the study at hand, the author provided a set of example sentences with variations in preposition use that were found through explorations of the Corpus of Global Web-based English (GloWbE) corpora to be reasonably frequent yet diverging in different varieties of English. Here, this was predominantly based on UK and US contexts, but information on the spread of the respective alternative uses was also provided for twenty different countries included in GloWbE. The extracted model sentences were provided to the teachers and acceptability ratings were obtained in a way that mimics real marking scenarios before showing the true corpus distribution and prompting them to provide another acceptability rating given the new, corpus-based information on real-world use. The picture that emerged shows that much policy work is needed in the area of using corpora in the classroom. No large-scale change of acceptability ratings could be observed due to teachers being presented with the corpus-based information. On the contrary, the ratings remained nearly unchanged, and some respondents entirely rejected any merit of frequency-based, descriptive approaches to correctness and acceptability. The author emphasises a need to expose practitioners more extensively to data of this kind and create awareness of the necessity to provide students with an appropriate, modern and ‘post-native’ teaching style to optimally prepare them for real-world language use. Work of the kind that is presented in this chapter is crucial since it helps shape further research to actively consider its policy implications, and it is a great example of where linguistics can most benefit from crossing boundaries, here into sociology, to maximise the impact and applicability of research.
In the final chapter, ‘Query a corpus in near-natural language: A human-friendly corpus query language not only for linguists’ (pp. 248–62), the flavour of innovation at hand changes yet again and diverges from World Englishes when Jiří Milička and Denisa Šebestová discuss their practical work developing a tool for easier implementation of corpus searches. This chapter is linguistics adjacent but highly relevant to corpus practitioners. It begins by describing accessibility issues to do with Corpus Query Language (CQL), a standardised query format widely used in corpus linguistics research. While it is a natural feature of any query format that a training phase is necessary to establish competency in formalising search queries and defining parameters, the authors provide a tool aimed at making CQL more accessible, particularly for users from other disciplines without prior training and users with dyslexia. The tool, Alpha translator, automatically converts near-natural language queries such as find lemma “check” followed by possessive pronoun, then follows lemma “privilege” (in Czech or English) into the most likely CQL syntax (here, this would be [lemma="check"][feats=".*Poss=Yes.*" & upos="PRON"][lemma="privilege"]). The authors further emphasise that their approach to designing this tool is rule-based as opposed to a machine-learning algorithm, but has actively been designed to be robust against misspellings and varying syntax. The rationale for using Alpha over Large Language Models (LLMs) is that Alpha does not hallucinate and can be integrated into larger LLM-based query construction processes as a reliable and replicable endpoint. The chapter further provides a host of examples for different prototypical corpus linguistics research areas and serves as the first English language tutorial for the tool. The majority of examples and testing procedures described in this chapter have been carried out on the InterCorp corpus, a multilingual collection of fiction texts in thirty-two languages with Czech as the pivot language. Overall, this rather technical chapter highlights the importance of democratising access to corpus linguistics tools through reliable, low-effort natural-language interfaces, and reflects recent broader efforts in the field to bridge the gap between corpus research and non-expert practitioners, e.g. in classroom settings (Crosthwaite Reference Crosthwaite2024).
Collectively, these ten chapters demonstrate the volume’s success in crossing boundaries – methodological, disciplinary and technological. It remains to be said that this volume offers a valuable blueprint for the continued evolution towards more inclusive, computationally sophisticated, interpretable and interdisciplinary approaches that one can hope would shape the next chapter for corpus linguistics.
