Form variation of pronominal it-clefts in written English

2.1 It-clefts

If a focus is pragmatically needed, e.g. to contrast a piece of information, it will usually be grammatically marked, either prosodically or morphosyntactically, or both (Lambrecht 1994: p. 64). The expression of this marking can eventually be conventionalised, and cleft constructions are a good example of such grammaticalisation.

English it-clefts consist of the pronoun it, followed by a form of the verb to be, a clefted constituent, and a complementiser, which introduces a relative clause that is attributed of the clefted phrase (Lambrecht 2001). It-clefts introduce two meaning parts: (i) a presupposition that the property in the clause following the complementiser holds of some entity; and (ii) an assertion (at-issue contribution) that this property holds of the entity denoted by the clefted constituent. We will thus call the clause following the complementiser the presupposed clause, following Prince (1978); the clefted constituent is also called the focussed constituent.

The literature on clefts has concentrated on several central questions of syntax and semantics. On the one hand, research has addressed the issue of the underlying syntactic structure of cleft sentences. On the other hand, given that the truth conditions between a cleft sentence and its corresponding simple clause do not differ, there has been discussion about the exact pragmatic contribution of using a cleft construction, especially with regard to information structure (Delin 1992).

Here, we focus on the form of the cleft construction itself, which offers a lot of variability within the frame defined above. We identify the following “moving parts” of it-clefts:

The example in (1) illustrates a contracted, nominative cleft with the complementiser that, whereas (2) is an uncontracted, accusative cleft with who as the complementiser.

Of these features of variation in clefts, Maier (2013, 2014) studies the case of the pronoun using corpora. Maier (2013) shows that the form of the pronoun is influenced not only by its position in the sentence, but also by its function, with variation related to the discourse medium: while written language prefers nominative-case pronouns, spoken English uses more accusatives. Maier (2014) accounts for the variation in case with information structural rationale, claiming that the nominative case of English pronouns in clefts was refunctionalised to signal emphasis (which he calls “focus”). We note that since he investigates both it-clefts and other related constructions together, the conclusion that it-clefts with nominal case pronouns are more emphatic than the accusative ones does not strictly follow from his statistical analysis.

The case of the pronoun in it-clefts is also mentioned by Patten (2010, 2012). Patten notes how the case of the pronoun is fundamental to the type of syntactic account: under an expletive account (in which the cleft is used to mark a sentence with the addition of a “dummy” element), the case of the postcopular element should be nominative, but this pattern “is not found in the dialect of the majority of speakers” (Patten 2010: p. 146, reporting a finding in Akmajian 1970). However, using the ICE-GB corpus (Nelson et al. 1998), Patten (2012) shows that clefted pronouns in the nominative are almost twice as frequent as those in the accusative (with a proviso of very low counts for either), and agrees with Akmajian (1970) in interpreting this as the preference for a prestige form, which gives an appearance of formality. This is further confirmed by a corpus study in Biber et al. (1999), where clefts with a nominative pronoun and who are frequent in fiction and news despite a general predominance of the accusative form. This is motivated claiming that nominative forms may be perceived to be more correct, possibly because of prescriptive rules moulded on those of Latin grammar.

As for complementisers, Collins (1991) finds that about two-thirds of it-clefts contain that, but most other wh-words are viable for the role. Cheshire et al. (2013) conduct an in-depth investigation of complementisers in restrictive relative clauses. They find that who is used almost exclusively in subject relative clauses. The choice of complementiser shows a strong variation in their data, though for younger speakers, most instances are realised with that. The complementiser that is the most general (lacking the animacy requirement of who). For some speakers, Cheshire et al. (2013) conclude that the subject preference of who is being grammaticalised as (additional) “topic” marking, in the sense that referents relativised with who are more salient (therefore, more available for re-mention or more “topic-worthy”) in subsequent discourse.

Looking at their distribution across registers, Collins (1991) notes that clefts outnumber pseudo-clefts in writing, while the opposite happens in speaking. Within spoken genres, it-clefts are more used in prepared monologues than in dialogues, further pointing in a direction of higher formality. However, internal features such as the case of the pronoun or the complementiser of the cleft are not analysed.

We are not aware of studies that systematically investigate multiple axes of variability in English it-clefts. In this study, we will consider these variants within a theory of emphasis marking.

2.2 Emphasis

We define emphasis as a linguistic means chosen by a speaker to direct a hearer’s attention, for example to help single out an intended referent (Zimmermann 2008). This concept has been inconsistently referred to throughout the pragmatics literature, being at times likened to prosodic stress (e.g. Heath 2018) and often used interchangeably with other under-defined or theory-specific terms such as “prominence” or “salience”. Clefts of all kinds are emphatic in the sense that, as reflected by their prosodic prominence (Lambrecht 2000), they identify the clefted element as something that the speaker considers especially worthy of attention, because of its status as new or contrastive information. In addition to the grammaticalised emphatic function of clefts, syntactic or morphological (or indeed prosodic) cues exist and can be used by the speaker to indicate particular emphasis.

Computer-mediated communication, being written, does not mark emphasis prosodically. However, it makes available a large range of other emphasis markers, that have been shown to take information structural roles. McAteer (1992) studies how the alteration of typeface can mark information focus in a way completely akin to intonational signals. Some of these are standard in writing and have conventionalised functions, such as punctuation or paragraphing, others are not, like all caps or boldface. These functions are included by McAteer (1992), along with clefting, in a set of “foregrounding strategies” that a writer can employ to mark contrast and emphasis. In more recent work on writing in social media, Heath (2018) identifies non-standard capitalisation as a marker of focus and emphasis, while Scott (2015) analyses hashtags as topic markers.

To study the variable encoding of the cleft construction and its relation to emphasis marking, we annotate these types of emphasis phenomena in our data:

Social media has been shown to innovate written language systematically, with a big influence coming from spoken language: for example, spelling on Twitter appears to follow phonological variations relatively frequently (Eisenstein 2013a). Non-standard emphasis markers are, likewise, more widespread in social media than in standard written language.

2.3 Formality

Another dimension which interacts with emphasis, in particular in social media, is formality. Heylighen and Dewaele (1999) define expressivity of language (which is another reshaping of emphasis in that it corresponds to linguistic genres with high detail and emotionality) as well as formality in terms of the orthogonal variability axes fuzziness and context-dependence. In this view, context-dependence refers to the amount of inference that has to be made in order to interpret an expression, fuzziness refers to the opposite of semantic precision of expression. Formal style will tend to minimise fuzziness and context-dependence, while expressive language is precise (not fuzzy) but highly context-dependent.

In this account, expressivity is a linear combination of precision (-fuzziness) and context-dependence: for example, a proper name will be much less fuzzy than a pronoun, which will in turn be highly context-dependent. Likewise, the two complementisers will vary along the fuzziness axis: who, with its [+animate] trait, includes more information than that, and being more precise it will be more formal and more expressive. Its use could thus correlate with the avoidance of contraction in the cleft sentence (which increases its phonological weight and thus the formality of the register: see Kjellmer (1997) for extensive examples). Similarly, the use of emphasis markers indicates increased expressiveness and could thus be associated with uncontracted clefts, or with a more precise form of the complementiser like who.

Further, it is known that social media are a highly informal medium, often reflecting features frequently found in spoken conversations more than written texts (cf. Tagliamonte and Denis 2008; Storrer 2013; Scheffler 2017). This places social media texts in the middle of the “conceptual orality” spectrum (Koch and Oesterreicher 1985) which moves along an axis of medium (spoken to written) and one of formality (informal to formal). Since emphasis by itself is also an effect of markedness, the specific use of a formal variant in social media would constitute a marked case and thus increase the overall emphasis of the utterance. In the following, we investigate the use of formal and emphatic variants of it-clefts in Twitter data.

3.1 First data collection and annotation

Tweets including at least one cleft were collected on 25th to 29th March 2019 using TAGS (Hawksey 2016). The queries were url-encoded to target specific tweets without excluding other factors such as emphasis markers or punctuation, and excluded retweets.^[4] The features targeted in the searches are listed in (4): contraction, case of the pronoun, and complementiser. This means all it-clefts introduced with or without a contraction (it’s ^[5] and it is), with a third person singular pronoun in either nominative or accusative case, and that or who as a complementiser, followed by a presupposed clause of any length.

All tweets using a cleft in the context of a quote from sacred scriptures (e.g. “But thou shalt remember the LORD thy God: for it is he that giveth thee power to get wealth”) were manually removed from the dataset because they are representative of a different register of language. Many of these appeared multiple times as “quotes of the day”. This left the dataset with a total of 798 tweets that conform to our constraints.^[6]

The data was manually annotated for the following features:

Number of referents: 1 for clefts wherein only one referent was mentioned (e.g. “it’s he that needs to go”), 2 for clefts involving two referents (e.g. “it’s she that called him”).

Role of the clefted element: Subject (e.g. “it is he that assisted Castro”) versus Object (e.g. “it is he that I will respond to”). This annotation only concerns clefts with two referents (i.e. clefts with both a subject and an object).

Emphasis: presence of emphasis markers: see Section 2.2.

Cleft: text of the cleft(s), isolated from the rest of the tweet.

45 tweets (5.6% of the total) contained multiple clefts. Only the first cleft of every tweet was considered in the following analysis.

3.2 Twitter data analysis

The first analysis targets the full dataset and some of the features annotated in the tweets. The basic distribution of the three main factors listed in (4) (contraction, pronoun case and complementiser) is shown in Figure 1. It can be gathered that in general clefts in English-language tweets are more often contracted and contain an accusative pronoun with the complementiser who. Moreover, more clefts with masculine than with feminine pronouns are produced (489 vs 309, 61.3 vs 38.7%; this ratio is not visualised).

Figure 1:

Ratios of the three features in the data set.

Figure 2 shows the distribution of all eight combinations of the main features considered. It seems evident that the features are not distributed randomly: three combinations dominate the picture. About equally frequent, they together account for about 3/4 of the it-cleft occurrences. These are exemplified in (6):^[7]

Figure 2:

Distribution of the eight feature combinations in the data. The outline of the bars distinguishes uncontracted (continuous line) from contracted (segmented line) clefts, the color of the bar distinguishes the case of the pronoun (nominative = red, accusative = blue), and the intensity of the color distinguishes the complementiser (who = full color, that = transparency).

In the following, we statistically evaluate the correlations between the individual features. Given that no direction of effect can be discerned, chi-squared tests^[8] were used to test for the associations and Cramér’s V for their strength. Cohen’s (1988) guidelines were used to evaluate the strength of the effects, with 0.1 < V < 0.3 indicating a small, 0.3 < V < 0.5 a medium, and V > 0.5 a large effect size. We report only significant effects, except where noted otherwise.

The case of the pronoun is associated with both the use of contraction ( χ 2 = 230.02 , p < 0.0001 , V = 0.54 : large effect size) and the presence of emphasis markers ( χ 2 = 12.47 , p < 0.001 , V = 0.13 : small effect size). This means that clefts are more contracted with accusative pronouns and less with nominative pronouns, and that emphasis markers co-occur significantly more often with nominative case pronouns. These effects are shown in Figure 3.

Figure 3:

Associations of pronoun case with contraction and emphasis. Figures 3–6 show the co-occurrence of two features, represented on the x axis and through colour. The y axis shows the proportion of each combination.

Figure 4:

Associations of complementisers with case and contraction.

Figure 5:

Distribution of case and syntactic role in clefts with two referents. The difference between subjects and objects is not statistically significant.

Figure 6:

Associations of complementisers with the syntactic role of the clefted element.

The choice of complementiser is associated with the case of the pronoun: nominative pronouns are used more frequently with who ( χ 2 = 78.31 , p < 0.0001 , V = 0.31 : medium effect size). Moreover, the complementiser who co-occurs significantly more with uncontracted clefts ( χ 2 = 19.46 , p < 0.0001 , V = 0.16 : small effect size). The two effects are visualised in Figure 4.

3.3 Syntactic role of clefted pronouns

Subsetting the data to the 148 clefts involving two referents, an obvious prevalence of clefts of the subject can be seen (112 vs 36, 75.7 vs 24.3%). While the subject is naturally a topical element in the information structure of an utterance, the object often provides new information, making it less necessary to mark explicitly with a focus.

In structural positions, the pronoun case is determined by the syntactic role the pronoun occupies (with nominative for subjects, and accusative for objects). However, this is not found in the clefted positions in our data. There is no significant association of the syntactic role of the clefted element with the case of the pronoun ( χ 2 = 3.67 , p = 0.06 ). As shown in Figure 5, both subjects and objects are more likely to appear in accusative case in our Twitter data.

No significant association of syntactic role with the use of contraction in the cleft was found ( χ 2 = 0.004 , p = 0.95 ), nor with the presence of emphasis markers ( χ 2 = 1.5 , p = 0.22 ). On the other hand, the syntactic role is associated with the complementiser: clefts of the subject co-occur significantly more with who rather than that ( χ 2 = 15.47 , p < 0.0001 , V = 0.32 : medium effect size). This is visualised in Figure 6.

3.4 Replication

The associations above were confirmed with a second, not manually annotated batch of 616 tweets mined on eighth May 2019. Table 1 compares percentages in the two datasets. Three of the chi-squared and Cramér’s V tests could be replicated on the non-annotated data, yielding similar results (see Table 2). This replication shows that the results from our first analysis are stable across data samples.

3.5 Comparison between Twitter and general web English

These associations were then recalculated on the results of corpus searches in the iWeb Corpus (Davies 2018), a 14-billion-word corpus of written English data from 94k+ systematically-chosen websites.

Within the 4056 pronominal it-clefts found in the iWeb corpus, all associations seen in the Twitter data are replicated (with p < 0.001 ). Case is associated with contraction ( χ 2 = 783.24 , V = 0.44 ) and with complementiser choice ( χ 2 = 362.28 , V = 0.3 ), and contraction is associated with complementiser choice ( χ 2 = 89.15 , V = 0.15 ). While the effects are more significant, likely because of the dimensions of the corpus, Cramér’s V figures show slightly smaller effects (especially for the association of contraction and pronoun case). Note that same direction of the effects will be confirmed through logistic regression.

The Twitter and iWeb results were compared through chi-squared tests in a 2 × 16 contingency table (including gender among the variations), computing p-values by Monte Carlo simulation (Hope 1968) because of the low counts for some of the structures. A significant difference was found ( χ 2 = 1827.5 , p < 0.001 ) between how Twitter and general-purpose (web) English use clefts of different types. Figure 7 shows how Twitter users adopt more accusative pronouns and contractions, while all-purpose language in iWeb attests at a more formal level avoiding contraction, as dictated by some style manuals, e.g. the Wikipedia Manual of Style (Wikipedia contributors 2019). Nominative pronouns are used more outside of Twitter, mostly because of the very high frequency of the structure It is he who (forming 77% of all nominative-pronoun structures). Complementisers are used similarly across genres.

Figure 7:

Comparison of the iWeb corpus with Twitter. The features on the right-hand side of each plot are those more heavily prescribed in writing (nominative case, avoidance of contraction, and who).

A generalised mixed-effects logistic regression (Bates et al. 2015) was fitted to predict the “genre” binary (Twitter vs iWeb) from the features of the clefts, i.e. case, contraction and complementiser, with a random effect for the gender of the pronoun (not assumed to display systematic variation across genres). All predictors were two-level factors. Model comparison was conducted via asymptotic likelihood ratio tests between a model without interactions, one adding the two-way interactions, and one additionally including the three-way interaction. Maximal random effect structure was used when supported by the data (as recommended by Barr et al. 2013): the random effects were successively simplified, chosen by lowest variance, until convergence was reached. The final model included a random intercept for pronoun gender. Adding the two-way interactions improved the fit ( p < 0.001 ), while the three-way interaction did not ( p = 0.45 ). The model results are reported in Table 3.

The model confirms that in Twitter clefts use fewer nominative pronouns ( p = 0.015 ) and are more frequently contracted ( p < 0.001 ). An interaction shows that clefts in which nominative pronouns and who co-occur appear less in Twitter than in iWeb ( p = 0.009 ).