Stability and Fluidity in Syntactic Variation World-Wide

Abstract

What are the scope and limits of syntactic variation within and across varieties of English? To address this question, we investigate well-known syntactic variation between the s-genitive (Mr Barnsley’s management) and the of-genitive (the management of Mr Barnsley) in nine varieties of English. We specifically gauge the stability of constraints on this variation by analyzing a richly annotated dataset spanning 10,558 interchangeable genitives from nine components of the International Corpus of English. Regression modeling indicates that constraints such as possessor animacy, constituent length, final sibilancy of the possessor, as well as the effect of medium (spoken vs. written) as a language-external factor differ in strength across varieties. The language-internal constraints, however, never change effect direction. We conclude that the probabilistic grammar fueling genitive variation is surprisingly stable overall, but does exhibit some fluidity along the lines of a distinction between English as a native language (ENL) and English as a second language (ESL) varieties: those constraints that tend to favor s-genitive usage tend to be weakened in ESL varieties.

Keywords

variation corpus syntax probability mixed-effect model genitive alternation World Englishes

1. Introduction

Establishing the scope of grammatical differences between varieties of English around the world in a comparative perspective is a quite popular research topic in the recent literature (e.g., Gries & Deshors 2015; Kortmann & Wolk 2012; Szmrecsanyi & Kortmann 2009). We contribute to this scholarship by exploring the constraints on syntactic variation in a comparatively large and typologically diverse sample of varieties of English. As a case study we specifically investigate the so-called “genitive alternation” between the s-genitive, as in (1), and the of-genitive, as in (2).

(1) Parliament also removed additional powers granted to him last year to tackle [the country]_possessor’s [economic crisis]_possessum under these powers. (ICE-SIN, s2b-001)

(2) Cement is one of the core raw materials for a developing country like India and it plays a vital role in the [economic growth]_possessum of [the country]_possessor. (ICE-IND, w2a-031)

The two variants differ in their ordering of the possessor and possessum phrases, and also in an additional definite article preceding of-genitive possessums. In the spirit of recent work on the genitive alternation (e.g., Ehret, Wolk & Szmrecsanyi 2014; Grafmiller 2014; Shih et al. 2015), we adopt a variationist perspective (Labov 1972; Jankowski & Tagliamonte 2014) and restrict attention to genitive tokens where the alternative variant could have been used. Thus, for example, of-constructions as in “You know most of the lecturers” (ICE-SIN, s1a-001) are not considered, as a periphrasis with the s-genitive (*the lecturers’ most) is not possible.

Genitive variation is extremely well studied. Historically, the s-genitive, after having been increasingly replaced by the of-genitive in Middle English, made a comeback in Early Modern English (e.g., Thomas 1931). In Late Modern English, the frequency of the s-genitive increased further primarily thanks to an expansion to inanimate noun classes (Wolk et al. 2013). The relevant literature about the constraints that govern the choice between the s-genitive and the of-genitive is too voluminous to be reviewed here in much detail (see Rosenbach 2014 for an exhaustive overview). Suffice it to say that the determinants of genitive variation are numerous, multifactorial, and probabilistic. Constraints that are well known to influence genitive choice include, but are not limited to, possessor animacy (animate possessors favor the s-genitive), constituent length (long possessors favor the of-genitive and long possessums the s-genitive), and final sibilancy (a final sibilant in the possessor discourages the s-genitive). Crucially, however, none of these factors is deterministic: for example, animate possessors do favor the s-genitive, but if the possessor is long enough, the principle of end-weight may win out against possessor animacy. To come to terms with the multifactorial nature of genitive variation, analysts have long seen the need to use multivariate techniques for analyzing corpus data (Gries 2002), or to use experimental designs (Rosenbach 2005).

Research has uncovered interactions between language-internal constraints (e.g., animacy, length) and language-external factors such as genre, time, and variety (Hinrichs & Szmrecsanyi 2007; Wolk et al. 2013; Grafmiller 2014). However, as far as regional/geographic differences are concerned, analysts have tended to restrict attention to differences between British and American English, and we know next to nothing about genitive variation in the many other varieties of English spoken and written around the world—this is the gap in the literature that we seek to fill. In contrast to much previous research, this paper offers an analysis that is not solely focused on English as a native language (ENL) varieties, e.g., British, Irish, Canadian, and New Zealand English, but also considers a number of English as a second language (ESL) varieties such as Jamaican, Singapore, Indian, Philippine, and Hong Kong English.

That said, we would like to emphasize at the outset that our main interest does not lie in assessing the aptness of particular models of variety categorization and genesis (e.g., Kachru 1985; Schneider 2007), or for that matter in characterizing particular varieties of English. Rather, we are concerned in this paper, in a more typologically inspired and thus abstractive spirit, with the scope and limits of syntactic variation within and across varieties of English around the world. Our investigation is guided by two research questions:

To what extent do users of different varieties of English rely on the same or similar choice-making processes when it comes to choosing genitive variants?

Are cross-varietal differences random or can they be explained by variety type (ENL vs. ESL)?

On the theoretical plane, we commit to the notion that grammar is the “cognitive organization of one’s experience with language” (Bybee 2006:711) and apply the idea of a dynamic probabilistic grammar (e.g., Bybee & Hopper 2001; Gahl & Garnsey 2004) to the realm of variation across varieties. We specifically rely on the variation-centered, usage- and experience-based probabilistic grammar framework developed by Joan Bresnan and collaborators (e.g., Bresnan & Ford 2010). This framework makes three key assumptions:

(i) Grammatical variation is sensitive to multiple and sometimes conflicting probabilistic constraints. Such constraints influence linguistic choice-making in subtle ways that may remain invisible unless analyzed quantitatively.

(ii) Grammatical knowledge must have a probabilistic component, for the likelihood of finding a particular linguistic variant in a particular context in a corpus has been shown to correspond to the intuitions that speakers have about the acceptability of that particular variant, given the same context.

(iii) This probabilistic knowledge is derived in large part from language experience, and so is subtly—but fluidly—(re)constructed throughout speakers’ lifetimes.

What were our predictions and hypotheses prior to embarking on this study? It seems reasonable to assume that language users, whatever variety they speak or write, are subject to the same cognitive and processing constraints and are therefore prone to make overall similar syntactic choices. MacDonald (2013), for example, proposes a unified Production-Distribution-Comprehension (PDC) approach that explains how biases in language production lead to statistical patterns in production that language users implicitly learn and that subsequently guide comprehension. One of these biases in (incremental) language production is what MacDonald (2013:3) calls the “Easy First” principle: language users tend to place those constituents first that are comparatively easy to retrieve, so the execution of utterances can begin early.

Easy First is a more general account for well-known tendencies such as the principle of end-weight, according to which longer constituents tend to follow shorter constituents—shorter constituents are “easier,” hence the word order pattern (Behaghel 1909; Wasow 1997). We expect to see end-weight effects in the genitive alternation, i.e., a preference for the variant that places long constituents after short constituents (as illustrated in 3 and 4), across the board, given their presumably strong links to the design of the human speech production system.

(3) In [today]_possessor’s [retention-conscious climate of trying to keep numbers up and students in programmes]_possessum, I believe that this is a question that must be addressed [. . .] (ICE-CAN, w1b-021)

(4) but anyway it’s uhm again the [story]_possessum of [Bruny Surin and Donovan Bailey]_possessor and it’s the lead story in La Presse [. . .] (ICE-CAN, s1b-021)

Thanks to the experience-based nature of grammar, and the different input(s) that users of different varieties receive over their lifetime, it is likely that there are subtle differences in the relative influence of preferences, such as the principle of end-weight in particular speech communities (see Bresnan & Hay 2008 for discussion). In this connection, Szmrecsanyi et al. (2016) have coined the term “probabilistic indigenization,” which refers to the process through which probabilistic patterns of internal linguistic variation, such as end-weight effects, are reshaped by shifting usage frequencies in speakers of postcolonial varieties. We expected to see probabilistic indigenization effects not primarily with regard to constraints that are strongly processing-driven (such as, presumably, end-weight), but rather with constraints such as possessor animacy or register/medium differences (register conventions are, after all, social conventions).

Based on these considerations, we created a variationist dataset covering 10,558 interchangeable s- and of-genitives drawn from nine components of the International Corpus of English. A mixed effects logistic regression model (dependent variable: genitive choice) correctly predicts 94 percent of all genitive outcomes in the dataset and indicates that constraints such as possessor animacy, constituent length, final sibilancy of the possessor, as well as the effect of medium as a language-external factor differ in strength across varieties. Crucially, however, the language-internal constraints do not change effect direction, as predicted by Easy First (MacDonald 2013). This is another way of saying that users of English consistently place long constituents after short constituents, avoid the s-genitive when the possessor ends in a sibilant, and prefer the s-genitive when the possessor is animate. This overall stability notwithstanding, the subtle fluidity that we find does seem to distinguish ENL from ESL varieties: those constraints that tend to favor s-genitive usage tend to be downplayed in ESL varieties, and those that favor of-genitive usage tend to be strengthened. We will argue that in a synchronic perspective, this is an echo of the well-known second-language acquisition preference for analytic marking (an exploration of the alternative explanation that ESL varieties lag behind ENL varieties with regard to the diachronic drift toward more s-genitive usage is reserved for another occasion—see section 5).

The remainder of this paper is structured as follows. Section 2 describes the data source, the variable context, the constraints on genitive variation that we consider, and our methods. Section 3 presents the results. In section 4 we discuss our findings, and section 5 offers some concluding remarks.

2. Data and Methods

2.1. Corpus Data and Varieties Covered

The data for this study were extracted from the International Corpus of English (ICE) (Greenbaum 1991), a corpus family that is well suited for cross-varietal comparisons since all subcorpora follow the same design. They each contain 600,000 words of spoken and 400,000 words of written language and are further subdivided into a wide variety of text types to offer a balanced reflection of the varieties (Nelson 1996). The spoken component contains dialogues and monologues (face-to-face-conversations, phone calls, classroom lessons, broadcast discussions, etc.); the written component distinguishes non-printed (e.g., student writing) and printed (e.g., academic writing) texts (for an overview of all text categories in ICE, see The ICE Project 2009).

The present analysis explores nine ICE subcorpora, which represent (i) British, (ii) Irish, (iii) Canadian, (iv) New Zealand, (v) Jamaican, (vi) Singapore, (vii) Indian, (viii) Philippine, and (ix) Hong Kong English. Although this selection was motivated by a desire to cover a wide range of varieties, some existing corpora could not be included (i.e., African varieties, or ICE-US) due to limited availability or format issues at the time of study design.

The varieties can be classified as ENL, or L1 (i–iv), and ESL, or L2 (v–ix) varieties. This particular classification is the customary one adopted in well-known reference works (Crystal 2004; Kortmann & Szmrecsanyi 2004; Kortmann & Lunkenheimer 2012). We are aware that this simple distinction is rather crude, and that more fine-grained approaches could have been taken, such as distinguishing between low-contact varieties and high-contact varieties (Trudgill 2009), or taking more account of the distinction between language shift varieties such as Irish English and other varieties (Hickey 2010). Further, some varieties are difficult to place in either category. Singapore English, for example, is said to be on its way to becoming an ENL variety (Pakir 2001; Schneider 2007). It is also subject to debate whether Jamaican English is an ESL variety (we classify JamE as ESL since it is Jamaican Creole that is the native language of most of the population, not Jamaican English, which is taught in schools; Patrick 2004; Hinrichs 2006). In spite of the imperfections of the classification, an ENL-ESL distinction seems to offer the appropriate level of granularity regarding the goals of the present study. We note in passing that the ENL-ESL distinction also translates into the terminology of customary models such as Kachru’s (1985) Three Circle Model (where ENL corresponds to the Inner Circle and ESL to the Outer Circle), or Schneider’s (2007) Dynamic Model (where ENL varieties occupy stage five, and ESL varieties occupy earlier stages), but we again hasten to add that the present study is not primarily concerned with models of variety classification or evolution.

Unlike other syntactic phenomena, genitive constructions are textually relatively frequent. A pilot study indicated that 10 percent of the texts in ICE yield enough observations for reliable statistical analysis, so a sample was created that contains text one, text eleven, text twenty-one, etc. The selection proportionately reflects all text categories. After automatic extraction of all genitive markers (i.e., of, ’s, and word-final s’), false positives were manually discarded, which resulted in a sample of 10,558 interchangeable genitives.

2.2. Defining the Variable Context

Following previous studies (e.g., Rosenbach 2002, 2014; Wolk et al. 2013), we considered as interchangeable any genitive token that did not fall into one of the following categories: appositive genitives, classifying genitives, double genitives, idiomatic/fixed genitives, partitive genitives, and any genitives with indefinite possessums.

In appositive genitives, the of-phrase is a post-modification whose head is co-referential with the head of the preceding noun phrase and usually describes it further (Biber et al. 1999). The expression in (5), for example, is not about a group supporting a US envoy’s idea (this reading would be interchangeable), but the of-phrase describes the idea that this group supports (i.e., sending a US envoy to Northern Ireland).

(5) The group supports the idea of a special US envoy being sent to Northern Ireland. (ICE-IRE, w2c-001)

Classifying genitives do not specify a possessum as a specific entity but express to which class it belongs. Example (6) shows the difference.

(6) England’s chairman of selectors, Ted Dexter, has bowed to the wishes of Botham who, rightly enough, is playing the king in Jack and the Beanstalk, a part, it appears, that might have been written into the old children’s story for him. (ICE-NZ, w2e-001)

While the first genitive, England’s chairman of selectors, denotes a specific person, the second one, the old children’s story, does not refer to a specific story of a specific group of old children, but specifies the class of story (i.e., a story for children).

Double genitives contain two genitive markers at once, as in (7). Idiomatic genitives are fixed patterns that only occur in one form, e.g., (8). Finally, partitive genitives express a measurement of some sort (e.g., time, distance, or value; see Biber et al. 1999) as in (9).

(7) A painting of Pete’s also forms part of this type. (ICE-JA, s2b-041)

(8) Florists could lose out on sales opportunities this Valentine’s Day [. . .] (ICE-JA, w2c-001)

(9) He’s calculated that a dollar’s worth of trees planted in nineteen forty-seven would produce sixty dollars worth of logs and lumber in nineteen ninety-five. (ICE-CAN, s2b-031)

Further, all of-genitives that did not contain a definite possessum were considered categorical and were therefore excluded. The reason for this is that in s-genitives, the clitic ’s has a determiner function, which results in definite possessums in all s-genitives. For of-genitives to be comparable to that, an additional definite article is needed (recall examples 1 and 2). Cases like a major focus of our neurosurgical practice in (10) were thus excluded. Bare plurals, like Infections of the CNS in (10), were also excluded for the same reasons.

(10) Infections of the CNS will certainly remain a major focus of our neurosurgical practice. (ICE-HK, w2a-021)

By defining interchangeability in this way, we assume that the norm of what constitutes the choice context is comparable across varieties. We believe this to be a valid assumption for various reasons. First, the assumption of a constant interchangeability norm is a crucial prerequisite for variationist/probabilistic analysis and enables researchers to see the big picture. If we assumed different norms, varieties would not be comparable. Second, careful manual screening of all cases in both the choice context and the categorical context did not reveal any patterns that suggest different norms. Further, our regression model (see section 2.4) obtained very good diagnostics, which lets us confidently attribute the variation to the various language-internal and language-external constraints that we study.

2.3. Constraints and Annotation

The extraction of interchangeable genitive occurrences from the corpus material was accomplished in three steps: (i) automatic extraction of all text units that contain one of the genitive markers of, ’s , or word-final s’; (ii) automatic filtering according to lexical, part-of-speech, and grammatical constraints; (iii) manual checking to correct errors in the automatic filtering process. After that, possessor and possessum phrases of each genitive case were manually annotated and their nominal heads were extracted. Finally, all cases were annotated for the language-internal and language-external predictors presented in the following sections: animacy, final sibilancy, givenness, thematicity, overall frequency, length, type-token ratio, medium, and variety.

Animacy of the possessor is arguably the most important constraint in the genitive alternation (e.g., Rosenbach 2005). It is so important that it is used in many prescriptive grammars of English (e.g., Murphy 2012) as a rule on how to use English genitives (s-genitive if possessor is animate, otherwise of-genitive). Previous research, however, has demonstrated that the animacy constraint can be overpowered by syntactic weight (Rosenbach 2005), that it is subject to diachronic change (Wolk et al. 2013), and that its strength can differ by variety (Hinrichs & Szmrecsanyi 2007; Hundt & Szmrecsanyi 2012).

The annotation of animacy was performed semiautomatically. Every possessor head’s animacy status was automatically classified and manually corrected where necessary.¹ After initially using a fivefold classification following Wolk et al.’s (2013) approach (i.e., distinguishing animate, collective, temporal, locative, and inanimate entities), the classification was simplified to a binary distinction between animate and inanimate entities due to a high degree of correlation between the more fine-grained animacy categories and several other predictors.

Table 1 shows that in our data inanimate possessors are predominantly realized as of-genitives. Animate possessors, on the other hand, are almost equally likely to occur in the s- as in of-genitive.

Table 1.

Genitive Choice and Possessor Animacy

Possessor animacy	s-genitives	of-genitives
Inanimate	743 (11.4%)	5780 (88.6%)
Animate	1869 (46.0%)	2195 (54.0%)

If a sibilant (i.e., [s], [z], [ʃ], [tʃ], [ʒ], or [dʒ]) is present at the end of the possessor phrase, as in (11), s-genitive usage is less likely for articulatory reasons (Zwicky 1987).

(11) The paradox’s conclusion [. . .] (ICE-IND, w2b-021)

Table 2 shows that in the dataset under investigation, s-genitives occur only in about 10 percent of the cases where the possessor phrase ends in a sibilant as opposed to about 30 percent if there is no final sibilant. Final sibilancy was annotated automatically by referring to the CMU Pronunciation Dictionary.² If a word could not be found in the dictionary, the annotation relied on orthography.

Table 2.

Genitive Choice and Final Sibilancy of Possessor

Final sibilancy	s-genitives	of-genitives
Absent	2331 (29.6%)	5535 (70.4%)
Present	279 (10.4%)	2413 (89.6%)

Givenness, also sometimes referred to as information status (e.g., Jankowski 2013), captures whether or not information has been mentioned in the preceding context. Many previous studies have shown that given possessor heads favor s-genitive use, but not always significantly so (e.g., Grafmiller 2014). For this study, we coded a possessor as given if the lemma of the possessor head occurred anywhere in the respective corpus text prior to the observed genitive token. The annotation of givenness was performed automatically by a script that searched the previous context of each genitive case and used a lemma list to determine if the lemma in question had been mentioned before.³ In the present dataset (see Table 3), given possessor heads are more likely to be realized as s-genitives (28.9 percent) than discourse-new possessors (22.4 percent).

Table 3.

Genitive Choice and Givenness

Possessor givenness	s-genitives	of-genitives
New	1531 (22.4%)	5299 (77.6%)
Given	1079 (28.9%)	2649 (71.1%)

Thematicity (Osselton 1988) reflects the degree to which a possessor head constitutes a central topic of a corpus text. It is defined here as the frequency with which a possessor head occurs in the entire corpus text in question: if a possessor head is highly thematic, it is more likely to take the s-genitive (Hinrichs & Szmrecsanyi 2007). The total number of mentions was counted automatically and then normalized by the usual corpus text size of 2000 words. In our data, possessor heads in s-genitives are more thematic than possessor heads in of-genitives (Table 4).

Table 4.

Genitive Choice and Thematicity of Possessor (Mean Frequency per 2000 Words of Running Text)

Thematicity	s-genitives	of-genitives
Mean	10.8	9.1
Standard Deviation	12.8	12.5

To our knowledge, the overall frequency of the possessor head has not been included in genitive alternation research so far, but frequency has been shown to influence other grammatical choices (Hilpert 2008). To determine how often possessor heads are used overall, we automatically counted their occurrence in the Corpus of Global Web-based English (GloWbE corpus), a large-scale 1.9 billion word corpus of online English, which samples a multitude of different varieties and mirrors the ICE family in its 60/40 division of less formal and more formal texts (Davies & Fuchs 2015). Frequencies were retrieved from the respective components of GloWbE that represent the variety in question—possessor heads from Indian English, for example, were counted in the Indian component of GloWbE.

The summary in Table 5 shows that the possessor heads in of-genitives are more frequent on average than the possessor heads in s-genitives. It also shows a high standard deviation, which indicates that the frequency values deviate from the mean value quite a bit.

Table 5.

Genitive Choice and Overall Frequency of the Possessor Head (Frequency per Million Words in the GloWbE Corpus)

Head frequency	s-genitive	of-genitive
Mean	215.6	232.9
Standard Deviation	460.4	614.6

The length of the possessor and possessum is predicted to influence genitive choice along the lines of the principle of end-weight, which was first described by Behaghel (1909) and which is defined as “the tendency for long and complex elements to be placed towards the end of a clause” (Biber et al. 1999). In the genitive alternation, this translates to relatively long possessors, such as the Chinese Government in (12), favoring the of-genitive, and relatively long possessums, such as potential medical uses in (13), favoring the s-genitive (Rosenbach 2002, 2005; Wolk et al. 2013).

(12) the [power ]_possessum of [the Chinese Government]_possessor (ICE-HK, s1a-021)

(13) [laser]_possessor’s [potential medical uses]_possessum (ICE-IND, w2b-031)

There are different ways of measuring syntactic weight (e.g., number of characters, number of words, number of syntactic nodes), which all correlate very highly (Rosenbach 2014:227); following Wolk et al. (2013), we chose to use the number of orthographic characters. Whitespaces and special characters (e.g., hyphens) were not counted. In this study, we include separate measures of the length of possessor and possessum (e.g., Hinrichs & Szmrecsanyi 2007), rather than a ratio of the two (e.g., Bresnan & Ford 2010). Table 6 shows that possessors tend to be shorter in s-genitives and longer in of-genitives. Possessums show the reverse pattern (Table 7).

Table 6.

Genitive Choice and Possessor Length (Number of Orthographic Characters)

Possessor length	s-genitives	of-genitives
Mean	8.9	18.2
Standard Deviation	4.5	17.4

Table 7.

Genitive Choice and Possessum Length (Number of Orthographic Characters)

Possessum length	s-genitives	of-genitives
Mean	15.0	9.6
Standard Deviation	15.6	5.7

Type-token ratio, a measure of lexical density, represents how many unique words occur in a given text relative the total number of words in that text. Grafmiller (2014) shows that lexically dense environments favor the use of s-genitives. Since type-token ratio is very sensitive to overall text length, it was measured in the immediate one-hundred-word environment of the genitive instance in question. Table 8 shows that the average type-token ratio is slightly higher in s-genitives than in of-genitives.

Table 8.

Genitive Choice and Type-Token Ratio (Type-Token Ratio in the Immediate Environment of ±50 Words)

Type-token ratio	s-genitives	of-genitives
Mean	0.82	0.81
Standard Deviation	0.07	0.07

The literature also reports effects that language-external factors (such as medium and variety) have on genitive choice. Grafmiller (2014), for example, found that spoken texts favor s-genitives more than written texts (with the exception of press texts), and Hundt and Szmrecsanyi (2012) found that in different varieties (i.e., early British English and early New Zealand English), possessor animacy varies in strength. The present study focuses on variety, but also controls for medium (i.e., spoken vs. written language). It seeks to understand how genitive distributions differ across varieties and how variety interacts with the language-internal constraints discussed above.

As can be seen in Table 9, there are no obvious differences in genitive frequency between spoken and written texts when aggregated over varieties. Regarding variety differences, Table 10 shows that s-genitive rates range from 16.3 percent in Jamaican English to 34.7 percent in Canadian English.

Table 9.

Genitive Choice and Medium

Medium	s-genitives	of-genitives
Written	1569 (25.3%)	4632 (74.7%)
Spoken	1041 (23.9%)	3316 (76.1%)

Table 10.

Genitive Choice and Variety

Variety	s-genitives	of -genitives
Great Britain	284 (23.3%)	937 (76.7%)
Canada	380 (34.7%)	714 (65.3%)
Ireland	264 (26.5%)	731 (73.5%)
New Zealand	327 (25.5%)	957 (74.5%)
Singapore	310 (30.2%)	717 (69.8%)
Jamaica	173 (16.3%)	889 (83.7%)
Hong Kong	318 (27%)	859 (73%)
India	247 (18.5%)	1088 (81.5%)
Philippines	307 (22.5%)	1056 (77.5%)
Total	2610 (24.7%)	7948 (75.3%)

2.4. Data Analysis

In order to statistically model genitive choice in varieties of English, we utilized mixed-effects logistic regression modeling, which models multifactorial phenomena that are simultaneously influenced by multiple constraints. While some constraints (see section 2.3) favor the s-genitive (e.g., animate possessors), others favor the of-genitive (e.g., the presence of a final sibilant). Without controlling for multiple factors at once and estimating the contributions of each individual predictor while holding other predictors constant, it is hard to pinpoint which role individual constraints play. Additionally, there might be interactions between the constraints, such that the influence of one constraint might change depending on another constraint. The present investigation is particularly interested in how well-studied constraints (such as animacy and length) interact with variety (e.g., is the animacy constraint stronger in certain varieties than in others?). By using logistic regression modeling, we can both estimate the contributions of the individual constraints and account for possible interactions. The resulting model provides coefficients for every constraint and every interaction, and these reflect the degree to which the constraints influence genitive choice.

In order to improve the estimation of these coefficients and the generalizability of the model, we included random effects, which have been shown to make the estimation of the constraints that are of primary interest (fixed effects) more accurate (Gries 2015). Random effects capture the influence of constraints that are particular to the sample, i.e., not repeatable, or not of primary interest to the analysis (Gelman & Hill 2007; Baayen 2008:241). For example, idiolectal preferences are particular to the sample since a different sample from another corpus would most likely sample different language users. Other constraints that were modeled as random effects are the ICE-specific text categories “monologue,” “dialogue,” “printed,” and “non-printed” (Nelson 1996), as well as the head lemmas of the possessor and possessum phrases. Idiolectal differences were modeled as nested into text category (referred to as “Text category/Language user” in Table 11), accounting for the structure of ICE, which does not sample individual language users in more than one text category. All of these constraints were modeled as random intercepts.

Table 11.

Variance Accounted for by Random Effects

Group	Variance
Possessor head lemma	3.86
Possessum head lemma	2.12
Text category/language user	1.53

The model selection process followed the guidelines in Zuur et al. (2009) and Gries (2015), i.e., proceeding from a full model and gradually deleting unnecessary components, first in the random effects structure and then in the fixed effects structure. Model diagnostics were monitored in the process. The remaining coefficients were validated using a bootstrapping approach as outlined in Baayen (2008:283). In this approach, one hundred random subsets of the data are taken, and the same regression model is fit to the subset each time. The coefficients will be slightly different every time, which enables us to estimate how variable the coefficients are and thus deduce a range in which we are 95 percent sure the true value of the respective coefficients lies. This bootstrap analysis showed that these ranges never include zero, which means that we can be confident that our results are not due to chance. All calculations were performed with R (R Core Team 2015) and the lme4 package (Bates et al. 2014).

The resulting minimal adequate regression model classifies 94.14 percent of all cases correctly (baseline: 75.32 percent of-genitives). The concordance statistic C, which ranges between 0.5 (random classification) and 1 (perfect classification) and which is independent of the baseline, is 0.981. C-values above 0.8 indicate that the model is good enough to explain the variation patterns in the data (Tagliamonte & Baayen 2012:156). The model shows medium multicollinearity (κ = 18.06), which is well below the “potentially harmful” threshold of thirty (Baayen 2008:182). Thus we may assume that the estimation of a given predictor’s coefficient is not distorted by the influence of any other predictor.

Table 11 shows the significant random effects in the resulting regression model. The higher the variance value, the more do members of the respective group exhibit a bias toward one of the genitive variants. Possessor head lemma has a higher variance than possessum head lemma, which indicates that there are more possessor head lemmas with a bias toward either the s-genitive or the of-genitive than possessum head lemmas. Examples include temporal and locative possessor head lemmas like week and Jamaica, which tend to be realized as s-genitives, and possessor head lemmas like God and department, which are part of phrases like the kingdom of God or the head of the department, which tend to be realized as of-genitives. In the case of the possessum heads, it is job titles (e.g., director in director of administration, and president in president of committee) that favor the of-genitive, and it is often words that are linked to a prototypical relation of possession that tend toward s-genitives (e.g., hat in woman’s hat, and house in friend’s house). Language user idiosyncrasies nested into text category show less variance, which indicates that idiolectal variation is comparatively subtle.

3. Results

Table 12 shows the effects that the individual constraints have on genitive choice. The coefficients (column b) and the odds ratios are effect sizes, which reflect how the presence/absence of the constraints (e.g., the presence of a final sibilant) or, in the case of numerical predictors, a one-unit increase (e.g., +1 character in possessor length) changes the odds for s-genitive occurrence. If b is positive (e.g., 4.31 for the change of possessor animacy from inanimate to animate), the level change in question makes s-genitives more likely (e.g., animate possessors facilitate s-genitives); if it is negative, s-genitive use is less likely. The odds ratios show similar information, but have a lower bound at zero, and a value of one indicates an equal chance of either variant. If the odds ratio is greater than one, s-genitives are more likely than of-genitives; if it is between zero and one, they are less likely. Odds ratios can be interpreted as factors with which the odds of an s-genitive realization are increased or decreased given certain predictor changes. For example, the presence of a final sibilant in the possessor phrase changes s-genitive odds by a factor of 0.35, i.e., makes s-genitive usage only 0.35 times as likely in the presence of a final sibilant as compared to its absence.

Table 12.

Regression Coefficients and Odds Ratios of Individual Predictors in the Minimal Adequate Model

Predictor	b	Odds ratio	p
(Intercept)	−4.84	0.01	<.001
possessor animacy
inanimate → animate	4.31	74.19	<.001
possessor length (logged)
one-unit increase	−2.40	0.09	<.001
possessum length (logged)
one-unit increase	0.55	1.74	.026
final sibilancy
absent → present	−1.04	0.35	.002
possessor givenness
new → given	0.25	1.29	.017
medium
written → spoken	0.31	1.36	.446
possessor head frequency (logged)
one-unit increase	−0.11	0.90	<.001
possessor thematicity (logged)
one-unit increase	0.16	1.17	.003
type-token ratio
one-unit increase	2.83	17.01	<.001
variety
Great Britain → Canada	1.24	3.44	.010
Great Britain → Hong Kong	1.52	4.55	.002
Great Britain → New Zealand	1.30	3.67	.007
Great Britain → Philippines	1.16	3.20	.013
Great Britain → Singapore	1.02	2.76	.034

Note: Predictions are for the s-genitive. The baseline for the factor variety is British English.

We can see that possessor animacy makes a huge difference in genitive choice, with animate possessors increasing s-genitive odds by a factor of seventy-four vis-à-vis inanimate possessors.⁴ Possessor length is also highly significant, and every one-character increase in length makes s-genitives less likely. Possessum length affects genitive choice in the opposite direction: for every additional character an s-genitive outcome becomes more likely. With a negative coefficient for possessor length and a positive coefficient for possessum length, the predictions are in line with the principle of end-weight. The effects of final sibilancy and givenness are also as expected, with final sibilancy disfavoring and givenness favoring s-genitive use. Medium, as a main effect, does not make a significant difference in the model (p > .05), but was not excluded since it is part of a higher-order interaction (see Table 13). Thematicity and type-token ratio also increase s-genitive use and thus influence the alternation as expected. Possessor head frequency, however, slightly but significantly discourages an s-genitive outcome, which is unexpected. Finally, there are five significant level changes in variety, all of which favor s-genitives in comparison to the British English baseline. They are, in order of effect size, Hong Kong English, New Zealand English, Canadian English, Philippine English, and Singapore English. Indian English, Jamaican English, and Irish English do not significantly differ from British English as far as the overall probability of the s-genitive is concerned.

Table 13.

Regression Coefficients and Odds Ratios of Predictor Interactions in the Minimal Adequate Model

Interaction	b	Odds ratio	p
possessor animacy : variety
animate + Hong Kong	−0.88	0.41	.040
animate + New Zealand	−0.88	0.41	.034
animate + Philippines	−1.65	0.19	<.001
possessum length (logged) : variety
one-unit increase + Canada	0.65	1.92	.046
one-unit increase + Hong Kong	0.71	2.03	.034
one-unit increase + Ireland	1.10	3.02	.001
one-unit increase + Philippines	0.73	2.08	.017
one-unit increase + Singapore	0.70	2.01	.037
final sibilancy : variety
present + Hong Kong	−2.73	0.06	<.001
present + India	−1.31	0.27	.022
medium : variety
spoken + Hong Kong	−1.38	0.25	.014
spoken + Philippines	−1.83	0.16	.001

Note: Predictions are for the s-genitive. The baseline for the factor variety is British English. Non-significant factor levels are not shown.

Table 13 reports significant interaction terms involving variety. The interaction of possessor animacy and variety shows negative coefficients across all varieties, significantly and most strongly so for Philippine English, Hong Kong English, and New Zealand English. This indicates that the tendency to use the s-genitive with animate possessors is strongest in British English. On the other hand, while the main effect of possessum length, reflecting the constraint’s strength in British English, is barely significant, it is considerably stronger in Irish, Philippine, Hong Kong, Singapore, and Canadian English, indicating that this factor is rather weak in British English. Final sibilancy also shows stronger effects in other varieties, significantly so in Hong Kong and Indian English. Medium does not seem to make a difference in British English (main effect not significant), but a change from written to spoken language does make a significant difference in Philippine and Hong Kong English. Remaining interaction terms between variety and other predictors not listed in the table were not significant.

To further highlight the nature of the interaction terms in the model, we now move on to present effect plots, which show fitted values on a probability scale. Effect plots visually summarize information from the regression model by systematically altering the values of individual predictors (e.g., animacy and variety) while holding the values of all other predictors in the model constant (see Fox 2003 for a detailed explanation). The effects plots in Figures 1 to 4 provide a visual representation of the significant interactions in the regression model, reflecting how users of different varieties are expected to choose variants for typical genitive contexts that only differ in the levels of the predictors involved in the respective interaction.

Figure 1.

Effects Plot of the Interaction Between Variety and Possessor Animacy

Figure 2.

Effects Plot of the Interaction Between Variety and Possessum Length

Figure 3.

Effects Plot of the Interaction Between Variety and Final Sibilancy

Figure 4.

Effects Plot of the Interaction Between Variety and Medium

Figure 1 shows the interaction of variety and possessor animacy. The probability of s-genitive use in the animate condition is higher for the five varieties on the left side, i.e., the ENL varieties and Singapore English. Estimates for the other ESL varieties are considerably lower. Figure 2 shows the interaction of variety and possessum length. While users of Irish English make sure to use the s-genitive with very long possessums, users of the other varieties show a weaker tendency to do so. Users of British English and Indian English, on the other hand, seem unaffected even by very long possessums. Figure 3 shows the interaction of variety and final sibilancy. Users of Hong Kong and Indian English are very unlikely to use s-genitives in the presence of a final sibilant in the possessor phrase. Final sibilancy seems to make the least difference in Philippine English, where the estimates for “present” and “absent” are closest. Finally, Figure 4 shows the interaction of variety and medium. S-genitive predictions are slightly elevated in spoken Canadian, New Zealand, and British English, while medium does not seem to make a difference in Irish, Singapore, Jamaican, and Indian English. In Philippine and Hong Kong English, predictions for s-genitive use are clearly higher in written texts.

4. Discussion

Our analysis of the scope and limits of syntactic variation between the s-genitive and the of-genitive was guided by two questions: first, what is the extent to which users of different varieties of English rely on the same or similar choice-making processes when it comes to selecting genitive variants? Second, are cross-varietal differences random or can they be explained by variety type (ENL vs. EFL), based on a classification that was derived from standard reference works (Crystal 2004; Kortmann & Szmrecsanyi 2004; Kortmann & Lunkenheimer 2012)? With regard to the first question, we found that language-internal constraints in genitive variation are fairly stable across varieties, with degrees of fluidity that remain within certain limitations. As to the second question, we argued that our results point to a probabilistic distinction between ENL and EFL varieties, especially with regard to the effect of possessor animacy.

Recall that s-genitive proportions (see Table 10) are lower in some ESL varieties (Jamaican, Indian, and Philippine English) than in the ENL varieties we studied (British, New Zealand, Irish, and Canadian English). It turns out that similar to the ENL varieties, Singapore and Hong Kong English (which are ESL) also show high s-genitive rates, a finding that is robust under multivariate control in a regression analysis (not reported here) without interactions. How do we account for these frequency differentials? Given that language learners tend to avoid inflectional marking and prefer analyticity (Klein & Perdue 1997:311) and that languages with a high proportion of language learners tend to lose case marking (Bentz & Winter 2013), we would like to argue that users of ESL varieties tend to favor the more analytic, periphrastic, and thus more transparent of-genitive in contexts where ENL speakers might be more likely to use the s-genitive. Hence we find higher frequencies of of-genitives in ESL varieties—in other words, the of-genitive is a comparatively stronger default option in ESL varieties than in ENL varieties.

We move on to discussing the conditioning of genitive variants in multivariate analysis. The nature of the main effects conforms across the board with the Easy First principle (MacDonald 2013), according to which language users tend to place constituents that are more easily retrievable earlier for the sake of optimizing utterance planning. Indeed, we consistently find that shorter genitive constituents are placed first, and that animate, given, and highly thematic possessors, which are also easier to retrieve (MacDonald 2013:3–6), favor the s-genitive. There is one constraint in our analysis that prima facie violates Easy First: frequent possessors, which should be easier to recall from memory (MacDonald 2013:3), were predicted to favor the s-genitive, but in fact were found to discourage s-genitive use. We suspect that this may be due to the fact that proper noun possessors, as in (14), which are low in frequency, are mostly realized as s-genitives.

(14) In Chong Fung-yuen’s case, it might be said, the CFA has politely put its foot down. (ICE-HK, w2b-011)

So the probabilistic constraints we considered are overall well behaved, but how does our analysis shed light on the issue of stability versus fluidity in probabilistic grammars? The majority of language-internal constraints under study do not significantly interact with variety—and this is another way of saying that we are observing a good deal of cross-varietal stability. Users of English, whatever their regional and/or cultural background, choose genitives in similar ways, likely because they are subject to the same cognitive constraints in language production. This, of course, is not surprising. But the model also does show degrees of fluidity, which remain within the limitations dictated by an overall stable pattern of constituent ordering choice driven by utterance planning. The language-internal constraints that do interact with variety (i.e., possessor animacy, possessum length, and final sibilancy) do not affect genitive choice in different directions across varieties (e.g., animate possessors favoring the s-genitive in variety A, but favoring the of-genitive in variety B)—we merely find different degrees of strength in the same direction.

Psycholinguistically inspired proposals in the spirit of MacDonald’s (2013) PDC account do not easily explain these differences in effect strength. Relying on this account, one would have predicted that language users in ESL varieties show stronger utterance planning biases since in many cases they constitute language learners who use English alongside one or more other languages (Kachru 1985:12). Findings from previous research suggest that the memory-related mechanism of Easy First should have a stronger effect in L2 production than in L1 production since in L2 processing there is constant interference from the L1 language (MacWhinney 1997; Szmalec, Brysbaert & Duyck 2012:89). Therefore, one would have expected to find that language users of ESL varieties more often choose s-genitives, e.g., with animate possessors. What we do in fact find, however, is that ESL users, similar to their overall preference as represented in the main effects, more often choose the of-genitive with animate possessors. This trend is amply clear from Figure 1. Supplementary regression analysis (not reported here), in which the nine levels of the factor variety were conflated into two levels ENL vs. ESL, shows that the effect of possessor animacy is significantly weaker in ESL varieties than in ENL varieties. An Easy First-inspired explanation at any cost would have to suggest that in ESL varieties, animate possessors are not “easier” than inanimate possessors to the same degree as in ENL varieties—hence the weaker showing of possessor animacy in the regression models—but there is no evidence that we know of that would support this claim. The more likely explanation is that ESL users simply tend to weaken constraints that favor the s-genitive, and instead strengthen constraints that favor the of-genitive. In this connection, we note that the subtle probabilistic differences we have uncovered provide evidence for what Szmrecsanyi et al. (2016) term “probabilistic indigenization,” i.e., a gradual shift in usage patterns in postcolonial varieties. Since we observe differences in the strength of constraints that influence the choice between the two genitive variants, we conclude that probabilistic genitive grammars are subtly but measurably distinct from each other as well as, in the case of postcolonial varieties, from their input varieties.

We reiterate that the direction of the effect of language-internal constraints is stable across varieties: long constituents, for example, always follow—not precede—short constituents, etc. There was one predictor in the model, however, whose effect changed direction across varieties: medium. Its positive coefficient 0.31 in Table 12 and the slightly elevated estimates in Figure 4 for the spoken condition of some varieties indicate a slight, however not statistically significant, tendency toward greater s-genitive usage in spoken texts. In Philippine and Hong Kong English, on the other hand, the spoken medium in fact disfavors s-genitive usage when compared to the written medium. We suspect that in Philippine and Hong Kong English, which are less advanced on the evolutionary path of Schneider’s (2007) Dynamic Model, there is an even stronger need to go for the safer, more analytic of-genitive in spoken language, considering the online processing constraints that this medium entails. But be that as it may, it does of course stand to reason that the predictor that is most unstable in the analysis, medium, is one that is language-external in nature and thus defined culturally. In this light, the hypothesis that we should see instability primarily with predictors that are not strongly processing-driven is borne out by the data.

Lastly, it seems worth mentioning that those constraints we find to be subject to probabilistic indigenization across varieties—possessor animacy, final sibilancy, and possessum length—are well-known to be unstable in diachrony. As for possessor animacy, the literature reports that the s-genitive has come to be increasingly used with collective, locative, and temporal possessors in some written genres (Wolk et al. 2013; Ehret, Wolk & Szmrecsanyi 2014). Final sibilancy has been found to increasingly disfavor s-genitive use over time (Hinrichs & Szmrecsanyi 2007), and longer possessums are now, according to the literature, more likely to take the s-genitive than they used to (Wolk et al. 2013; Ehret, Wolk & Szmrecsanyi 2014). The bottom line is that, as so often in work on language variation and change, synchrony and diachrony mirror each other.

5. Conclusion

The big question that this paper has sought to address concerns the extent to which the grammar of English is stable or fluid on a global scale. By tackling this issue, our work comes under the remit of research on English as a world language, but we add variationist/probabilistic rigor and new theoretical twists to this line of scholarship. By way of a case study, we investigated well-known syntactic variation between the s-genitive and the of-genitive in nine varieties of English. To establish the (in-)stability of constraints on this variation, we investigated a richly annotated dataset consisting of 10,558 interchangeable genitives from nine components of the International Corpus of English (ICE). Analysis showed that constraints such as possessor animacy, constituent length, final sibilancy of the possessor, as well as the effect of medium (spoken vs. written) are somewhat unstable in a cross-variety perspective. The language-internal constraints, however, never change effect direction. We are thus led to conclude that the probabilistic grammar regulating genitive variation in varieties of English is surprisingly stable overall. To the extent that we find fluidity and probabilistic indigenization, we see evidence for a distinction between ENL and ESL varieties: those constraints that tend to favor s-genitive usage tend to be weaker in ESL varieties than in ENL varieties.

As always, there are many ways in which the analysis reported in the present paper could be extended. For one thing, work is underway to annotate the dataset for additional language-internal constraints such as genitive relation, definiteness of the possessor, noun phrase expression type, and persistence/priming. We are currently also working on extracting genitives from the Corpus of Global Web-based English (GloWbE) (Davies & Fuchs 2015), with the goal of adding web-based language to the array of text types sampled in ICE. As regards the type of evidence we consider, corpus analysis has taken center stage in the present paper, but we are working on spot-checking the cognitive robustness of the corpus-derived probabilities via rating experiments along the lines of Bresnan and Ford (2010), who showed that language users’ acceptability (“naturalness”) intuitions about syntactic choices match probabilities as calculated in a corpus-based regression model.

Further research is encouraged to investigate predictions generated by our analysis. First, since we found possessor animacy to regulate genitive choice differently in ENL and ESL varieties, we predict that genitive use in EFL varieties (e.g., Edwards 2014) deviates from ENL genitive use even more than ESL varieties do. Second, we tentatively suggest that the strength of the possessor animacy constraint is proportional to how advanced a variety is according to Schneider’s (2007) Dynamic Model. Third, we predict that since animacy is a well-known constraint in a number of other syntactic alternations (e.g., the English dative alternation, relativizer choice, the active-passive alternation, etc.), it influences these alternations more strongly in ENL varieties than in ESL varieties: in ESL varieties, recipient animacy supposedly favors the prepositional dative less strongly, animate referents are less rigidly referred to by “who,” and animate objects are used less often in passive constructions. Last but not least, note that we took a strictly synchronic perspective in the present study. Research with a more explicitly diachronic twist based on longitudinal data sources is needed to investigate if the probabilistic preference for the of-genitive that we find in ESL varieties may (also) be due to historical patterns such that ESL varieties lag behind ENL varieties in the drift toward more s-genitive usage, for example thanks to a less enthusiastic expansion of the s-genitive to inanimate possessor categories.

Footnotes

Acknowledgements

We thank Joybrato Mukherjee and Dirk Speelman for valuable feedback.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding by the Research Foundation Flanders (FWO) is gratefully acknowledged (grant #G.0C59.13N).

Notes

Author Biographies

Benedikt Heller is a PhD student at KU Leuven (research group “Quantitative Lexicology and Variational Linguistics”) and investigates the genitive alternation in varieties of English. He obtained his master’s degree in 2013 from the University of Giessen (Germany).

Benedikt Szmrecsanyi holds a research professorship at the Department of Linguistics of KU Leuven (research group “Quantitative Lexicology and Variational Linguistics”). His research interests focus on variation studies (synchronic & diachronic), probabilistic grammar, language complexity, geolinguistics, and dialect typology. Recent books include Grammatical Variation in British English Dialects (2013, CUP), and Aggregating Dialectology, Typology, and Register Analysis (2014, ed. with Bernhard Wälchli, Mouton de Gruyter).

Jason Grafmiller is a post-doctoral researcher at KU Leuven (research group “Quantitative Lexicology and Variational Linguistics”). He obtained his PhD in 2013 from Stanford University.

References

Davies

Mark

Fuchs

Robert

. 2015. Expanding horizons in the study of World Englishes with the 1.9 billion word Global Web-based English Corpus (GloWbE). English World-Wide 36(1). 1-28.

Greenbaum

Sidney

. 1991. ICE: the International Corpus of English. English Today 7(4). 3-7.

The ICE Project. 2009. The Design of ICE Corpora @ ICE-corpora.net. International Corpus of English. http://ice-corpora.net/ice/design.htm (20 June 2016).

Baayen

R. Harald

. 2008. Analyzing linguistic data: A practical introduction to statistics using R. Cambridge: Cambridge University Press.

Bates

Douglas

Maechler

Martin

Bolker

Ben

Walker

Steven

. 2014. lme4: Linear mixed-effects models using Eigen and S4. Version 1.1-12. http://CRAN.R-project.org/package=lme4 (February 2016).

Behaghel

Otto

. 1909. Beziehungen zwischen Umfang und Reihenfolge von Satzgliedern. Indogermanische Forschungen 25(1). 110-142.

Bentz

Christian

Winter

Bodo

. 2013. Languages with more second language learners tend to lose nominal case. Language Dynamics and Change 3(1). 1-27.

Biber

Douglas

Johansson

Stig

Leech

Geoffrey

Conrad

Susan

Finegan

Edward

. 1999. Longman grammar of spoken and written English. Harlow: Longman.

Bresnan

Joan

Ford

Marilyn

. 2010. Predicting syntax: Processing dative constructions in American and Australian varieties of English. Language 86(1). 168-213.

10.

Bresnan

Joan

Hay

Jennifer

. 2008. Gradient grammar: An effect of animacy on the syntax of give in New Zealand and American English. Lingua 118(2). 245-259.

11.

Bybee

Joan

. 2006. From usage to grammar: The mind’s response to repetition. Language 82(4). 711-733.

12.

Bybee

Joan

Hoppe

Paulr

. 2001. Frequency and the emergence of linguistic structure. Amsterdam: Benjamins.

13.

Crystal

David

. 2004. The Cambridge encyclopedia of the English language. 2nd edn. Cambridge: Cambridge University Press.

14.

Denison

David

Scott

Alan K.

Börjars

Kersti

. 2010. The real distribution of the English “group genitive.” Studies in Language 34(3). 532-564.

15.

Edwards

Alison

. 2014. English in the Netherlands. Functions, forms and attitudes. Cambridge: University of Cambridge dissertation.

16.

Ehret

Katharina

Wolk

Christoph

Szmrecsanyi

Benedikt

. 2014. Quirky quadratures: On rhythm and weight as constraints on genitive variation in an unconventional data set. English Language and Linguistics 18(2). 263-303.

17.

Fellbaum

Christiane

. 1998. WordNet: An Electronic Lexical Database. Cambridge, MA: MIT Press.

18.

Fox

John

. 2003. Effect displays in R for generalised linear models. Journal of Statistical Software 8(15). 1-18.

19.

Gahl

Susanne

Garnsey

Susan

. 2004. Knowledge of grammar, knowledge of usage: Syntactic probabilities affect pronunciation variation. Language 80(4). 748-775.

20.

Gelman

Andrew

Hill

Jennifer

. 2007. Data analysis using regression and multilevel/hierarchical models. Cambridge: Cambridge University Press.

21.

Grafmiller

Jason

. 2014. Variation in English genitives across modality and genres. English Language and Linguistics 18(3). 471-496.

22.

Gries

Stefan Th

. 2002. Evidence in linguistics: Three approaches to genitives in English. In Brend

Ruth M.

Sullivan

William J.

Lommel

Arle R.

(eds.), LACUS Forum XXVIII: What constitutes evidence in linguistics, 17-31. Fullerton: LACUS.

23.

Gries

Stefan Th

. 2015. The most under-used method in corpus linguistics: Multi-level (and mixed-effects) models. Corpora 10(1). 95-125.

24.

Gries

Stefan Th.

Deshors

Sandra C.

2015. EFL and/vs. ESL? A multi-level regression modeling perspective on bridging the paradigm gap. International Journal of Learner Corpus Research 1(1). 130-159.

25.

Hickey

Raymond

. 2010. The Englishes of Ireland: Emergence and transportation. In Kirkpatrick

Andy

(ed.), The Routledge handbook of World Englishes, 76-95. London: Routledge.

26.

Hilpert

Martin

. 2008. The English comparative—Language structure and language use. English Language and Linguistics 12(3). 395-417.

27.

Hinrichs

Lars

. 2006. Codeswitching on the web. English and Jamaican Creole in e-mail communication. Amsterdam: John Benjamins.

28.

Hinrichs

Lars

Szmrecsanyi

Benedikt

. 2007. Recent changes in the function and frequency of Standard English genitive constructions: A multivariate analysis of tagged corpora. English Language and Linguistics 11(3). 437-474.

29.

Hundt

Marianne

Szmrecsanyi

Benedikt

. 2012. Animacy in early New Zealand English. English World-Wide 33(3). 241-263.

30.

Jaccard

James

. 2001. Interaction effects in logistic regression. Thousand Oaks, CA: Sage.

31.

Jankowski

Bridget Lynn

. 2013. A variationist approach to cross-register language variation and change. Toronto: University of Toronto dissertation. https://tspace.library.utoronto.ca/handle/1807/43604 (20 November 2015).

32.

Jankowski

Bridget L.

Tagliamonte

Sali A.

2014. On the genitive’s trail: Data and method from a sociolinguistic perspective. English Language and Linguistics 18(2). 305-329.

33.

Kachru

Braj B.

1985. Standards, codification and sociolinguistic realism: The English language in the Outer Circle. In Quirk

Randolph

Widdowson

Henry G.

(eds.), English in the world: Teaching and learning the language and literatures, 11-30. Cambridge: Cambridge University Press.

34.

Klein

Wolfgang

Perdue

Clive

. 1997. The basic variety (or: Couldn’t natural languages be much simpler?). Second Language Research 13(1). 301-347.

35.

Kortmann

Bernd

Lunkenheimer

Kerstin

(eds.). 2012. The Mouton world atlas of variation in English. Berlin: Mouton de Gruyter.

36.

Kortmann

Bernd

Szmrecsanyi

Benedikt

. 2004. Global synopsis: Morphological and syntactic variation in English. In Kortmann

Bernd

Schneider

Edgar

Burridge

Kate

Mesthrie

Rajend

Upton

Clive

(eds.), A handbook of varieties of English, vol. 2, 1142-1202. Berlin: Mouton de Gruyter.

37.

Kortmann

Bernd

Wolk

Christoph

. 2012. Morphosyntactic variation in the anglophone world: A global perspective. In Kortmann

Bernd

Lunkenheimer

Kerstin

(eds.), The Mouton world atlas of variation in English. Berlin: Mouton de Gruyter. http://www.degruyter.com/view/books/9783110280128/9783110280128.906/9783110280128.906.xml (13 February 2016).

38.

Labov

William

. 1972. Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.

39.

MacDonald

Maryellen C.

2013. How language production shapes language form and comprehension. Frontiers in Psychology 4(1). 1-16.

40.

MacWhinney

Brian

. 1997. Second language acquisition and the competition model. In De Groot

Anette M. B.

Kroll

Judit F.

(eds.), Tutorials in bilingualism: Psycholinguistic perspectives, 113-142. Mahwah, NJ: Lawrence Erlbaum.

41.

Murphy

Raymond

. 2012. English grammar in use. A self-study reference and practice book for intermediate learners of English. 4th edn. Cambridge: Cambridge University Press.

42.

Nelson

Gerald

. 1996. The design of the corpus. In Greenbaum

Sidney

(ed.), Comparing English worldwide: The International Corpus of English, 3-12. Oxford: Clarendon.

43.

Osselton

Noel

. 1988. Thematic genitives. In Nixon

Graham

Honey

John

(eds.), An historic tongue: Studies in English linguistics in memory of Barbara Strang, 138-144. London: Routledge.

44.

Pakir

Anne

. 2001. The voices of English-knowing bilinguals and the emergence of new epicentres. In Ooi

Vincent B. Y.

(ed.), Evolving identities: The English language in Singapore and Malaysia, 1-11. Singapore: Times Academic Press.

45.

Patrick

Peter L.

2004. Jamaican Creole: Morphology and syntax. In Kortmann

Bernd

Schneider

Edgar W.

Upton

Clive

Mesthrie

Rajend

Burridge

Kate

(eds.), A handbook of varieties of English, vol. 2, 407–438. Berlin: Mouton de Gruyter.

46.

R Core Team. 2015. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Version 3.2.3, retrieved December 2015 from http://www.R-project.org/.

47.

Rosenbach

Anette

. 2002. Genitive variation in English: Conceptual factors in synchronic and diachronic studies. Berlin: Mouton de Gruyter.

48.

Rosenbach

Anette

. 2005. Animacy versus weight as determinants of grammatical variation in English. Language 81(3). 613-644.

49.

Rosenbach

Anette

. 2014. English genitive variation—The state of the art. English Language and Linguistics 18(2). 215-262.

50.

Schneider

Edgar

. 2007. Postcolonial English: Varieties around the world. Cambridge: Cambridge University Press.

51.

Shih

Stephanie

Grafmiller

Jason

Futrell

Richard

Bresnan

Joan

. 2015. Rhythm’s role in genitive construction choice in spoken English. In Vogel

Ralf

Vijver

Ruben

(eds.), Rhythm in cognition and grammar. Berlin: Mouton de Gruyter. http://www.degruyter.com/view/books/9783110378092/9783110378092.207/9783110378092.207.xml (17 September 2015).

52.

Szmalec

Arnaud

Marc

Brysbaert

Wouter

Duyck

. 2012. Working memory and (second) language processing. In Altarriba

Jeanette

Isurin

Ludmila

(eds.), Memory, language, and bilingualism: Theoretical and applied approaches, 74-94. Cambridge: Cambridge University Press.

53.

Szmrecsanyi

Benedikt

Grafmiller

Jason

Heller

Benedikt

Röthlisberger

Melanie

. 2016. Around the world in three alternations: Modeling syntactic variation in varieties of English. English World-Wide 37(2). 109-137.

54.

Szmrecsanyi

Benedikt

Kortmann

Bernd

. 2009. The morphosyntax of varieties of English worldwide: A quantitative perspective. Lingua 119(11). 1643-1663.

55.

Tagliamonte

Sali

Baayen

Harald

. 2012. Models, forests and trees of York English: “Was/were” variation as a case study for statistical practice. Language Variation and Change 24(1). 135-178.

56.

Thomas

Russel

. 1931. Syntactical processes involved in the development of the adnominal periphrastic genitive in the English language. Ann Arbor: University of Michigan dissertation.

57.

Trudgill

Peter

. 2009. Vernacular universals and the sociolinguistic typology of English dialects. In Filppula

Markku

Klemola

Juhani

Paulasto

Heli

(eds.), Vernacular universals and language contacts: Evidence from varieties of English and beyond, 302-329. London: Routledge.

58.

Wasow

Thomas

. 1997. Remarks on grammatical weight. Language Variation and Change 9(1). 81-105.

59.

Wolk

Christoph

Bresnan

Joan

Rosenbach

Anette

Szmrecsanyi

Benedikt

. 2013. Dative and genitive variability in Late Modern English: Exploring cross-constructional variation and change. Diachronica 30(3). 382-419.

60.

Zuur

Alain F.

Ieno

Elena N.

Walker

Neil J.

Saveliev

Anatoly A.

Smith

Graham M.

2009. Mixed effects models and extensions in ecology with R. New York: Springer.

61.

Zwicky

Arnold M.

1987. Suppressing the Zs. Journal of Linguistics 23(1). 133-148.