The sociolinguistic study of English derivational morphology has rarely seemed to enjoy as much interest in published research as topics such as phonology, syntax, discourse, or even its closer cousin inflectional morphology. Ever since Romaine's (Reference Romaine1983) pioneering article on gender and the productivity of -ity and -ness, there have been sporadic shorter studies on the subject, but few book-length treatments, particularly those incorporating a diachronic perspective. A somewhat odd development in the subfield of historical word-formation has been that many of the best long-form studies have been unpublished (and sometimes hard-to-acquire) dissertations rather than monographs – e.g. Cowie (Reference Cowie1998), Anderson (Reference Anderson2000) and Lloyd (Reference Lloyd2005). Indeed, in closing a recently published piece on gender and the use of nominal suffixes (Palmer Reference Palmer2013: 19–20), I shone a light on the need for more (historical) sociolinguistic analyses of derivation:

This need, and so much more, has now been addressed by the recent publication of Tanja Säily's Sociolinguistic Variation in English Derivational Productivity – a significant contribution to our knowledge of historical sociolinguistics, morphology and statistically driven corpus methodologies.

The book addresses two major gaps in these fields: (a) How can researchers measure the productivity of suffixes in small corpora, since the standard measures, such as Baayen (Reference Baayen, Booij and van Marle1992), only seem to work well in large corpora? (b) Since sociolinguistic variables have been heretofore underexplored, how does an account of these variables help us better understand changes in derivational usage? To explore these questions, the book takes on the popular topic of suffixes -ity versus -ness. While the scope of affixes explored is narrow, the focus allows Säily to consider a larger range of data spanning multiple centuries – primarily the seventeenth, eighteenth and twentieth – and multiple corpora, including the Corpus of Early English Correspondence (CEEC), the Old Bailey Corpus (OBC), and the British National Corpus (BNC). The book argues for a more sophisticated and statistically rigorous account of productivity incorporating computational methods such as bootstrapping. It champions the value of collaborations between linguists and those in other fields, such as computer science and statistics, to further innovate methods in corpus linguistics. And it demonstrates the relevance of class and especially gender to understanding derivational use – primarily in discovering a ‘relatively stable gendered style’ (p. 233), as evidenced in the consistently low productivity of -ity in women's language over multiple centuries.

This book is not a monograph, but rather a published dissertation featuring both single-authored and collaborative studies, the latest from the series Mémoires de la Société Néophilologique de Helsinki. But even before being published, it's a dissertation whose content had already had a significant level of peer-review in different venues: e.g. chapter 9 appears, and chapter 10 will appear, in Corpus Linguistics and Linguistic Theory; chapter 8 features an article previously published in Literary and Linguistic Computing; and chapters 6, 7 and 11 are each included in different edited collections. Parts I (chapters 1–5) and III (chapters 12–15) present previously unpublished material, the former situating the various studies in the field and the latter teasing out some of the threads linking multiple chapters in the book. One problem with the inclusion of so many already published pieces in part II is the presence of repeated material throughout – many chapters recapitulate contextualizing information on methods, materials and previous studies that has been explained at length in earlier chapters. Though Säily provides a table (p. 6) to guide readers through topically related chapters, the sequencing of studies in the book is not made completely clear. Chapters 7 and 8, which do not address derivational morphology, need more framing, as they initially seem out of place in the book until later chapters build upon some of the conclusions in those chapters. Additionally, it's not immediately evident why chapter 6, which does address derivational morphology, is not clustered with the other productivity chapters (those after chapter 8). To take a closer look at the book's content, I will occasionally veer away from the book's chronology, quite often treating similarly themed chapters in groups.

The first part – consisting of chapters 1 through 5 – provides an overview of the book's primary questions, goals, materials and methods. Chapter 1 lays out central questions about five distinct but related topics: diachronic changes in the productivity of -ness and -ity; the problems of orthographic variation in small, historical corpora; the comparability of studies conducted on different corpora; the applicability of different productivity measures to sociolinguistic studies on corpora of different sizes; and the level of sophistication needed for tools used to study productivity (pp. 3–5). While chapters 2 and 3 provide thorough surveys of prior studies of productivity and historical sociolinguistics, respectively, chapter 4 provides a brief overview of the corpora explored. Chapter 5 advances the most substantive ideas in part I, offering a critique of commonly employed corpus linguistic methods. Säily persuasively argues that too many studies have relied on a bag-of-words test (e.g. chi-square or Fisher's exact), which falsely assumes that words are randomly distributed in a text and results in unjustifiable claims of statistical significance (p. 45). As a remedy, she advocates for three major improvements in corpus methods: (a) resampling – namely, permutation testing and bootstrapping, which involve computationally driven tests on random subcorpus samples of a larger corpus (pp. 47–9); (b) stricter corrections for p-values, due to multiple hypothesis testing on the same corpus (pp. 49–51); and (c) better visualization in modeling productivity, including innovations such as bean plots and type accumulation curves (pp. 51–9). Despite the strong case for such methodological improvements, on rare occasions the explication here is too hasty in its conclusions. For example, the curve in figure 5.4 (p. 56), which shows a statistically significant difference in type frequencies of -ity for two subperiods, is claimed to ‘indicate that the productivity of -ity increases over time’ (p. 54). But an increase in type frequency on its own merely suggests a possible increase in productivity; such a strong claim would be better backed by additional evidence (e.g. increasing hapaxes, increasing neologisms, increasing suffix transparency, etc.). Elsewhere, in this chapter and in later chapters, the text is typically much more cautious and nuanced in its claims and discussion of data.

Two studies from part II, presented in chapters 7 and 8, do not address productivity. But they do include vital information about corpus methods and part-of-speech frequencies that turn out in later chapters to be valuable for understanding sociolinguistic differences in the productivity of suffixes in nominals. Cleverly titled ‘CEECing the baseline’, chapter 7 primarily focuses on the question of single-genre corpus reliability and homogeneity over a period of time. Along with Säily, co-authors Jefrey Lijffijt and Terttu Nevalainen examine lexical frequencies to illustrate the continuity of the seventeenth-century portion of the CEEC, even after changes in the lexicon following the English Civil War. While the study does not sufficiently explain why lexical similarity should be the primary basis for establishing homogeneity/continuity – and not, say, morphosyntactic patterns or discourse conventions – it provides the book's clearest discussion and argument for carefully chosen statistical tests and stricter standards for p-values (pp. 98–105). In collaboration with Nevalainen and Harri Siirtola, Säily turns to the topic of the reliability of part-of-speech tagging in the Parsed Corpus of Early English Correspondence (PCEEC) in chapter 8, titled ‘Variation in noun and pronoun frequencies in a sociohistorical corpus of English’. While the chapter is unable to offer firm conclusions about the stability of the PCEEC over time or the reliability of its tagging, it does offer important caveats about the impact of corpus annotation on results, the problems of outliers, and the comparability of historical and Present-day English (PDE) studies based on differently tagged corpora. Most significant for the book is its finding that there is a ‘gendered style’ in noun and pronoun use within the historical data of the PCEEC that reflects PDE findings – namely, ‘women consistently us[e] more pronouns than men, while men use more nouns than women’ (p. 142).

The question of gender-based variation, and its relationship to measuring productivity, is the focus of chapters 6 and 9. The former chapter, ‘Comparing type counts: The case of women, men and -ity in early English letters’, is the result of a collaboration with a computer scientist, Jukka Suomela. A key observation is that women exhibit a lower, statistically significant accumulation of -ity types in the seventeenth-century portion of the CEEC than do men (p. 78). But the main achievement in this chapter is the successful presentation of an open-source tool to model the variation-dependent productivity of an item in a specific corpus. This computational program produces a type-accumulation curve with bands that represent different levels of statistical significance (increasingly smaller ranges of p-values). The genius of this graphic is that it allows the researcher to plot different subcorpora filtered by sociolinguistic variable – e.g. ‘gender, domicile, social rank, and time period’ (p. 79) – to enable easier comparisons of the potential productivity of a form among different social groups, text types and time periods. This methodology helps to solve a major problem for those of us in the field: the fact that the virtual gold standard in measuring productivity – Baayen's hapax-based models, counts of one-word occurrences of a form in vast stretches of text (Baayen Reference Baayen, Booij and van Marle1992) – does not work well for small corpora. Unfortunately, small corpora are sometimes all that is available for those who work in earlier historical periods. In fact, one of the best presentations on the limitations of Baayen's models for historical sociolinguistic research appears in chapter 9, ‘Variation in morphological productivity in the BNC’. Here Säily puts forth a strong case for cautiously using, or even in some cases avoiding, hapax-based measures when comparing corpora from different social subgroups (pp. 171–2). Her resistance emerges from evidence that ‘the number of hapaxes’ does not ‘grow linearly with the number of suffix tokens’ in a corpus (p. 171), even though Baayen's hapax measures assume linear growth. So, using Baayen's measure, a subcorpus with a greater number of suffix tokens, such as the set of men's data from the CEEC, would inaccurately show undervalued productivity when compared to a subcorpus with fewer tokens, such as the subset of texts written by women (for whom we commonly have fewer textual records).

To fill in some of the historical and methodological holes identified by her earlier work, Säily digs deeper into questions about variation in the productivity of -ness and -ity in chapter 10. Titled ‘Sociolinguistic variation in morphological productivity in eighteenth-century English’, it appropriately aims to bridge the historical findings on these suffixes from her work on the seventeenth century in the CEEC (chapter 6) and the twentieth century in the BNC (chapter 9). The chapter compares many more variables in its graphs than in previous chapters, considering not only gender but also various classes/professions and twenty-year subperiods. For -ness, Säily finds that ‘there are no significant differences in terms of gender, social rank or time period when productivity is measured in terms of type frequency as a function of the number of running words’, corroborating the stability of -ness productivity seen in the seventeenth and twentieth centuries (p. 187). However, when types are measured as a function of suffix tokens rather than total running words in the corpus, there is a class difference: clergy show high productivity with -ness while royalty show low productivity (p. 188). Another oddity is that the eighteenth-century portion of the CEEC is the only study in the book not to show a gender difference in the productivity of -ity. In other periods, and even in the trial proceedings of the OBC from the same period, Säily finds that women show significantly less type accumulation of -ity than do men. She convincingly attributes this anomaly to the make-up of the corpus, since the eighteenth-century portion of the CEEC underrepresents lower-class informants and perhaps overrepresents highly literate men and women, both of whom shared a similar writing style (p. 195).

This issue becomes a focus of a follow-up study, ‘Change or variation? Productivity of the suffixes -ness and -ity’ (chapter 11), which subjects the eighteenth-century data from the CEEC to even more scrutiny. Considering the data from the perspective of particular letter writers communicating with particular addressees, Säily finds that the professional class leads the increasing productivity of -ity during the period. And while class is found to be more significant than gender, it seems that the variable of addressee has the greatest impact on -ity’s productivity: professionals of both genders evidence a ‘wish to impress and amuse their friends by using the suffix’ (pp. 225–6). The chapter raises an important question that researchers need to consider further: is change in productivity better understood as a change in the language, or rather a change in style? While the chapter does not offer a definitive answer in the case of -ity in eighteenth-century letters, it rightly argues that the question should be tested further by exploring trends in a wider range of genres (p. 226).

Part III contains four short chapters to conclude the book. Chapter 13 on methods is the longest and most confident, putting forth an excellent case for using the book's resampling methodology to study productivity in a wider range of corpora and subcorpora than previous methods have allowed. Furthermore, Säily (p. 244) also acknowledges the applicability of these methods to other linguistic variables and the uptake that her approach is already receiving in the field – e.g. Gardner (Reference Gardner2013). The other chapters, however, are perhaps too short, at times not fully addressing the questions raised in the book, and at other times not championing enough the relevance of its findings to larger questions in the field of English linguistics. In addressing the five major research questions of the book, chapter 12 understates the methodological problems of spelling variation in historical corpus studies, concluding that it is a ‘minor issue’ because her methods ‘do not attempt to track the frequency of the base. . . or. . . all hapax legomena’ (p. 234). But even if one tracks affixes, especially in studies of corpora from before 1600, there is considerable (and sometimes surprising) spelling variation to account for in those affixes. And in small corpora, where even a few rare forms can matter a whole lot in accounts of productivity, hand-sorting becomes a necessity that may impede more computationally driven approaches. Chapter 14 summarizes the book's major findings on the social variation of the historical productivity of -ness and -ity, but it should have elaborated further their significance to larger questions in the field. If women have been mostly lagging behind men in the productivity of -ity historically, even when the productivity of the suffix has been on the rise, how does this square with our understanding of language change in the past and in PDE, where women have typically been found to lead non-standard changes? Are these changes in -ity’s productivity thus an effect of the historical development of the standard – an area of study which has largely ignored the role of derivational morphology? Or does the history of -ity depend significantly on particular non-standard varieties that need further investigation? And since -ness and -ity are only a very small part of a much larger derivational puzzle, what other forms should now be investigated to complete a clearer picture of these processes?

The good news is that Säily's text goes a long way to provide both results and methods to inspire others to find answers to these questions. Written with consistently clear prose and providing ample, well-illustrated data, the book is a major contribution to the fields of historical sociolinguistics, corpus linguistics and word-formation. Moreover, it is a brilliant example of the value of collaboration among researchers from diverse disciplines, combining insights and innovations from linguistics, statistics, computer science and graphic visualization studies. Such interdisciplinary collaboration can serve as a model for other linguists, who should feel further encouragement to look outside our field for additional ideas – especially when we confront many of our most fascinating, but somewhat nebulous linguistic concepts, such as morphological productivity.