Kristian Rusten's Referential Null Subjects in Early English provides significant evidence of the bidirectional relationship between historical linguistics (here, historical syntax) and diachronic corpora: it shows how the theory of historical syntax can be the basis for and the starting point of diachronic corpora investigations, but also how quantitative studies can trigger new directions for modern approaches to historical syntax. Rusten's study is a development of his University of Bergen PhD dissertation (Reference Rusten2015) and succeeds perfectly in testing all major hypotheses on the status of null subjects in Old English (OE) and their development. Most importantly, Rusten manages to present a clear empirical analysis of null subjects in English, as well as of various relevant approaches according to several models, which serve as a stable basis for sophisticated quantitative research. The theoretical discussion and analysis of relevant examples and the main characteristics of null subjects combine coherently with the empirical surveys presented in the book. The well-prepared structure of the book, the well-presented research questions and methodology, the clear short summaries in various parts and the detailed discussion of all results and of their theoretical implications support a clear presentation of Rusten's main proposal that no dialect of OE had null subjects of the type that pro-drop languages have, but null subjects in OE and their development are related to a remnant of an earlier productive system and are licensed by argument ellipsis.
Rusten's study can been seen as a model study for corpus-based investigations that also rely on state-of-the-art statistical methods: Rusten provides an in-depth quantitative analysis of 217,239 overt and null referential pronominal subjects (referential subjects that ‘are left unexpressed not due to conjunction reduction’, p. 16) that are attested in 509 OE, Middle English (ME) and Early Modern English (eModE) texts. Rusten clearly states that he investigates the referential subjects of OE that are not compatible with the rules of Present-day English (PDE) and excludes expletive null subjects, cases of conjunction reduction and subjectless imperatives from his investigation. He further excludes structures with the verb hatan ‘be called’ that contain an empty subject relative. The major aim of the corpus-based study is an analysis of the morphosyntactic and pragmatic characteristics of the non-expression of referential subjects in OE and an evaluation of the role that linguistic and extralinguistic factors play in (dis)allowing null subjects in OE. The author examines not only the dialectal factor but also other potential factors, such as the period of composition (early vs late OE), the translation status and the genre. With regard to linguistic parameters, the study investigates the role played by verbal agreement (the contrast between third-person and first- or second-person agreement) and by antecedent relations (according to topicality, accessibility scale and givenness hierarchy analyses).
Chapter 1 (Introduction) presents the main research questions of the study and discusses aspects of previous research on null subjects in Old Germanic languages: Gothic, Old High German, Old Saxon, Old Swedish, Old Norwegian and Old Icelandic. The contrast between OE and the other ancient Germanic languages is very significant for any study of the history of null subjects in English: even though the availability of referential null subjects in all other Old Germanic languages has been recognized in various studies, there is no consensus on the status of the null subjects in OE. Rusten's previous studies (Rusten Reference Rusten2013, Reference Rusten2015) form the starting point of the book: he has already shown that null subjects were very rare in the majority of the OE texts that he has examined. Rusten has also rejected the hypothesis of a West Saxon – Anglian split (see e.g. Walkden Reference Walkden2014) with regard to the characteristics of referential null subjects. A major difference between his previous studies and the book under review concerns the restricted number of OE texts investigated in the previous studies.
Chapter 2 presents a quantitative approach to the question of whether OE was a pro-drop language. The study presents in detail the numbers and the status of null subjects in each text of the corpus, doing so separately for prose and poems. His overall conclusion is that null subjects are in reality ‘a highly infrequent phenomenon in OE prose’ (p. 40). On the contrary, the relative frequency of null subjects in 34 texts of OE poetry demonstrates that null subjects are more common in poetry than in prose and can justify views supporting an analysis of OE as a pro-drop language. Rusten's main aim in chapter 2 is to test van Gelderen's (Reference Gelderen2013) hypothesis that OE is a ‘genuine’ pro-drop language and Walkden's (Reference Walkden2014) hypothesis of a dialectal split of OE in terms of null subjects. He also discusses examples of sentences with null subjects in Modern Norwegian, for instance, a non-pro-drop language – but not frequencies of null subjects in PDE. It will be of significance to show whether texts written in PDE, a non-pro-drop language without any doubt, display frequencies of null subjects similar to OE poems or OE prose texts, and to discuss in general the frequencies of null subjects in PDE. On the other hand, I find the discussion of various authors’ analysis of low relative frequencies (under 1%) as illustrating performance errors extremely significant. If null subjects in OE are not due to the canonical pro-drop properties of the language, then the crucial question concerns the reasons for the availability of null subjects as attested in OE texts. Rusten's explanation is that null subjects of OE are probably ‘remnants’ of the pro-drop system of prehistoric OE and Proto-Germanic.
In chapters 3 and 4, Rusten further develops the argument that OE is not a pro-drop language on the basis of in-depth quantitative studies of both linguistic and extralinguistic variables. Chapter 3 examines Walkden's hypothesis (Reference Walkden2014) that Anglian OE was a partial pro-drop language and tests possible statistical correlations between null subjects and the variables of dialect, period, genre and translation status. Rusten proposes that the observed variation can be attributed to genre variation and/or a diachronic decline of referential null subjects and a contrast between early and late OE texts, as well as possible influence from Latin in translated texts. Correlations between the above variables and null subjects may further weaken the dialect-split hypothesis. Therefore, Rusten uses robust statistical measures, generalized logistic regression modeling and a random forest of conditional inference trees to statistically evaluate the dialect-split hypothesis. The sophisticated statistical analyses presented in the second part of chapter 3 falsify not only Walkden's hypothesis of a dialect split, but also the hypothesis of translation effect and Latin influence and a hypothesis of a contrast between early and late OE texts. The contrast between poetry and prose appears to be the most salient extralinguistic variable on the basis of the robust statistical analyses, as well.
Chapter 4 investigates the morphosyntactic characteristics of null subjects in OE following a quantitative methodology. The author examines the referential null subjects of OE in relation to the clause type, person and number features and the position of the finite verb. Rusten uses a generalized mixed-effects logistic regression model and argues that the distribution of null subjects in OE, in contrast to ‘canonical’ pro-drop languages, shows a main–subordinate clause asymmetry. The author also investigates the role of person and number. However, he does not relate this factor to the question of the role of translation effects or the proposals of a contrast between third person and first and second person in the case of syntactic transfer and loan syntax (see Lavidas Reference Lavidas, Gianollo, Jäger and Penka2014; Lavidas & Tsimpli Reference Lavidas and Tsimpli2019). Following studies on characteristics of L2 grammars and the transfer from L1 to L2 (Interpretability Hypothesis: Tsimpli Reference Tsimpli, Liceras, Zobl and Goodluck2003 and her subsequent work), it has been suggested that translated texts, and contact-influenced grammars in general, share characteristics with L2 grammars and show a contrast (for instance, in languages that have clitics) between third-person clitics (which encode values of uninterpretable features only) and first- and second-person clitics (which carry an interpretable feature of person) (Lavidas Reference Lavidas, Gianollo, Jäger and Penka2014). On the other hand, for Rusten, the contrast between third-person and first- and second-person referential null subjects provides evidence in favor of a non-‘canonical’ pro-drop language. We should note, however, that possible contact-induced characteristics should not be seen as characteristics of the ‘source’ (or L1) language that have been directly transferred to the ‘target’ (or L2) language but as reflecting a contrast of the type third-person vs first- or second-person referential null subject (or other types of contrast, such as definite vs indefinite articles) according their (non-)interpretable features and the probability of their transfer into the ‘target’ (L2) language. On the basis of sophisticated statistical modeling, Rusten claims the correlations between referential null subjects of OE and the linguistic variables of clause type, person-number and the position of the finite verb are weak when both factors of individual texts and genre are taken into account. For this reason, the author argues that the above results should be seen as supporting the idea that null subjects of Old English are a linguistic ‘residue’: ‘this is a type of “residue” that surfaces very unevenly in different genres and texts while displaying only very tenuous associations with other variables’ (p. 120).
Chapter 5 attempts to answer the question of the licensing of null subjects in OE, starting again with the relevant studies of van Gelderen and Walkden. The chapter reveals how ‘claims put forward in central works couched in linguistic theory’ can be ‘subject to empirical testing’ (p. 121). Rusten's aim is also to evaluate the degree to which Sigurðsson's (Reference Sigurðsson1993) approach to antecedent relations can hold for the OE data. The author suggests that there are two types of counterevidence against Sigurðsson's predictions in the data of OE: (i) not all referential null subjects in non-verb-initial clauses have an overt antecedent, and (ii) not all antecedents of embedded and non-verb-initial referential null subjects are realized by NPs. In chapter 5, Rusten also offers long discussions of qualitative data, as in the case of example 12 (p. 129) from MS D of the Anglo-Saxon Chronicle. On the basis of a close analysis, he rejects the hypothesis that the null subject of this example is an impersonal arbitrary null subject that can be treated as a different type of null subject. All in all, Rusten proposes that referential null subjects in OE are ‘not categorically dependent on having an overt NP antecedent in the preceding discourse’ (p. 131). Moreover, he argues that the data of OE include a considerable number of referential null subjects that are not likely to have the degree of accessibility and topicality required by the models of Ariel (Reference Ariel1988 and subsequent work), among others. In this respect, Rusten concludes that the above cognitive and discourse models cannot adequately explain the status of referential null subjects in OE. Accordingly, quantitative and qualitative data show that referential null subjects in OE do not behave as would be expected if subject drop were due to its being in focus or topical. For the author, the above observation means that referential null subjects are not an active syntactic property in OE.
On pages 173–9, Rusten makes a new proposal. He suggests that null subjects in OE reflect a form of argument ellipsis, following Duguine's (Reference Duguine2014) approach to ellipsis. Accordingly, it is not necessary for Rusten to postulate any important change in referential null subjects that could justify differences between OE and PDE, for instance. The short section dedicated to this proposal does not include any discussion of the implications of such analysis for Rusten's main claim that null subjects in OE are residuals. The question that arises is whether we may see this type of development of null subjects as a case of refunctioning of an early characteristic – that is, whether early subject omission became a form of argument ellipsis in a later stage.
Chapter 6 examines the development of null subjects in ME and eModE. Rusten provides an overview of the distribution of null subjects according to period (early vs late), text and genre. In addition, he offers an overview of the morphosyntactic characteristics of null subjects in ME and eModE and models the examples of null subjects through a generalized mixed-effects logistic regression model. Rusten shows that null subjects are an option in ME poetry only to a very low degree (1.3%). His main conclusion is that the differences between the OE and the ME data cannot support the existence of a significant change from a system where referential null subjects were grammatical to a system where they are not. In this respect, the ME null subjects pattern along with OE null subjects: they are infrequent; they are most often attested in V1 clauses, with a third-person reference. We should note that the book contains long qualitative and philological discussions, besides statistical surveys and quantitative analyses. See, for instance, the discussion on the subjectless pray construction as a discourse or pragmatic marker in eModE on page 195. Examples with referential null subjects are attested more rarely in eModE than in ME (0.4%). Rusten observes that the differences in the frequencies of null subjects in eModE vs ME and in eModE vs OE are statistically significant but the effect sizes are negligible, which leads Rusten to consider the more sophisticated statistical surveys a necessity. According to the statistical analyses, there is a modest diachronic decline in the proportion of null subjects over time, which does not allow the author to revise the conclusions reached through the analysis of OE referential null subjects. Rusten argues that referential null subjects are attested at very infrequent and ‘comparatively stable’ proportions over time, from OE until eModE. The conclusion again disconfirms the suggestion of a parametric or typological shift in the history of English. Rusten, however, observes a change that concerns the contrast between third-person and first- and second-person subjects; the eModE data do not display this type of contrast. For him, this change is again not significant because of the low frequency of null subjects in OE and ME. On the other hand, if we take L2 studies into consideration, the difference in terms of contrast between third and non-third person can be seen as an example of transfer due to language contact that is active only in OE and ME but not in eModE. According to the author, only conjunct environments and verb-initial clauses – that is, factors that favor argument ellipsis – constitute favorable effects for referential null subjects in all examined periods (OE, ME, and eModE). Chapter 7 summarizes in a very clear way the main aspects of the proposed analysis and the main conclusions of the study.
Rusten has prepared a model study of how we can test various hypotheses following a quantitative methodology: he tested Walkden's dialect-split hypothesis – but he also added other extralinguistic (period, genre, translation status) and linguistic variables (the position of the finite verb, person and number features, clause type). He has clearly shown that all variables display very little influence on the status of referential null subjects in OE. In addition, he has argued that approaches that rely on inflectional morphology or Aboutness topicality as well as on the Accessibility scale and the Givenness hierarchy also fail to account for the characteristics of referential null subjects in OE. It is evident that further questions emerge for investigating other hypotheses of correlations and their significance for the OE data: for instance, the study has not examined the correlations between word orders and types of pro-drop, such as the case of a correlation between types of pro-drop and VSO orders (for the historical aspects of this correlation, see Lavidas Reference Lavidas and Tsangalidis2009, Reference Lavidas2013). A broad question also concerns the obvious significance of empirical studies in disconfirming theoretical hypotheses, and whether it is true that there are fewer cases in which quantitative and statistical in-depth analyses confirm theoretical analyses and hypotheses (especially in the case of historical surveys).
Rusten's book constitutes an example of how recent generative approaches can be compatible with early philological works, and how complete a study can be if it takes into consideration both modern linguistic and philological directions of research. The fact that the author makes the search parameters of his study available in the online supplement to the book contributes to the great validity of the research that can be easily replicated. Moreover, the annotated corpora were not constructed for the purposes of this study, which also ensures greater objectivity of the study. Rusten also gives the formulas he used to compute regression models. In addition, the book contains glosses, Present-day English translations and ID tags for all examples it includes. It could probably contain fewer abbreviations. For instance, it would be better if the author did not use abbreviations in the case of names of languages (other than English).
Rusten has succeeded in fulfilling the main aims of his study: he has offered us a large-scale quantitative corpus-based investigation of referential null subjects in OE and their development. In my view, he has also proven that one cannot conduct a historical linguistic investigation without diachronic corpora and that one cannot have quantitative analyses of the data derived from diachronic corpora without a stable basis of historical linguistic theory (theory of language change and approaches to historical syntax).
