The humanness factor

Much of the writing on the relative system in Early Modern and Modern English appears to take for granted that who is used for animate/human subjects in English. According to contemporary grammar books, who occurs with human (or personal) antecedents, and which occurs with nonhuman (or nonpersonal) antecedents (Quirk, 1957:97–98). Typical statements read as follows:

This so-called modern distinction for who as the most frequent form for human antecedents, as well as which with nonhuman antecedents, and zero mainly in nonsubject relatives (e.g., Quirk, 1957) would certainly be expected to operate in these 21st century data. Yet clearly it does not. Which is not used for “things” in any general sense because it barely occurs at all. However, what about who? The next step is to separate out the antecedents in the subject relative data according to semantic qualities of humanness and animacy. We coded separately for five different types: humans, as in (21), collective nouns that represented humans, as in (22), the frequent lexical item people, as in (23), inanimate “things” as in (24), and finally nonhuman animates (e.g., animals), a category that crosscuts the −human/+animate dimensions, as in (25), to determine what semantic distinction is being encoded.

If the “modern function” for who is in operation, we would thus expect to find a correlation with human subjects. Further, if humanness is the underlying semantic quality of relevance, humans should pattern independently of animals. If, on the other hand, animacy is the quality of relevance, then humans and animals should pattern similarly.

Table 6 shows the distribution of subject relative markers by the humanness of the antecedent separated by community. First, Table 6 shows that there is no lack of human subject relatives in our data. Indeed, these represent 65% of all the antecedents, a finding that is consistent with most studies of relative clauses. As we have seen, the frequency of who in these data is very low in general (Tables 2 and 3). Table 6 reveals that when who is used at all, it marks human subjects the vast majority of the time. Thus, for example 42/75 (56%) of the who tokens in Lowland Scotland are human; 45/65 (69%) in Northwest England, and 13/17 (77%) in Northern Ireland. However, when all the human subjects in the data are taken into account, it is readily apparent that there is no tendency for who to mark human subjects. Indeed, in every locality who is actually the least likely relative marker to mark the human antecedents – 13% in Lowland Scotland, 17% in Northwest England, and only 6% in Northern Ireland.

Distribution of relative markers (SUBJECT FUNCTION ONLY) by the animacy of the antecedent by community

Notice, however, that in Northwest England and Lowland Scotland who is used much more frequently with the specific lexical item people, as in (23). In these contexts, who is used at well over twice the frequency (33%, 40%) than it is used with human subjects (13%, 17%). Although the number of examples with people is small (N = 18 CMK; N = 42 MPT), the fact that who is so frequent for them suggests that it may be somewhat formulaic here – perhaps a lexicalized insertion in a highly restricted collocational context.

In fact, given the small number of who tokens overall (N = 130), it may be the case that there are other anomalies in its use that can provide further insight into its function in these communities. A closer look at every individual use of who in each of the communities proves revealing from a more sociolinguistic perspective.

In Lowland Scotland, who is used by less than half (44%) of the speakers (18/41). More tellingly, out a total of 58 tokens of who in this dialect, nearly 60% come from two speakers only. The remaining who forms are used once or twice each by the other speakers who use it. In Northwest England, who is used by only about half the speakers as well (20/43, 50%). Here, out of a total of 77 tokens, four speakers are prevalent users, but the rest use it less than five times each. Finally, in Northern Ireland, only 41% of the speakers use who (12/29). Here, out of a total of 25 tokens, one speaker produces a third of the tokens. In other words, in each community, use of who is not representative of community norms, but is circumscribed, either to specific lexical items like people, or to specific individuals. Interestingly, these are the individuals with a bit more education, and who are implicated in local affairs.¹⁸

In sum, despite the fact that WH forms have been attested, even championed, as being the marker for subject relatives for the last two centuries, their use is limited in frequency, surprisingly infrequent in its most favored function, lexically marked, and highly circumscribed sociolinguistically in these data. Such qualitative and quantitative patterns of use bear the characteristics of a prestige form imposed from the outside, rather than a productive part of the community's vernacular grammar.

Summary of distributional analysis

Given these overall distributions, we are now in a position to return to the first question posed at the beginning of this article. How far has the WH relative marker penetrated spoken data, in particular, vernacular dialects? First, the sheer lack of WH forms in these data confirms Romaine's suggestion that “the infiltration of WH into the relative system can be seen as completed in the modern written language … but it has not really affected the spoken language” (Romaine, 1982a:212). Indeed, the only WH form that occurs with any frequency at all is who.

But the question is: Why would this be true? Is it simply that these materials represent a particular dialect region whose inventory of relative markers is somehow different from other vernaculars (Northern Scots, Irish English, and English; see e.g., Montgomery, 1989:114)? Alternatively, can the results be more usefully explained when put into context with the history and development of the English relative markers more generally? We now consider the first possibility – that the propensity for that in these data is related to geographical distribution.

A dialect explanation?

Of course, we cannot avoid the possibility that the distribution of relative markers in these materials is unique to their geographic locale. After all, each community is positioned in a particularly well-defined region in northern Britain – not simply geographically, but historically, socially, and culturally, as well. And indeed, there are considerable reports from the literature to suggest the use of WH words is regionally delimited in Britain. For example, the historical record attests to the fact that northern dialects have favored that as a relative since at least as far back as Middle English (Jones, 1972:140; Nevalainen & Raumolin-Brunberg, 2002:112). Further, Jones (1972:140) claimed that northern British dialects have continued to favor the pæt/that relative over time in comparison to their southern counterparts. Romaine's large-scale analysis (1982a:222) confirmed this in observing that “certain non-standard varieties of English, like Scots, have lagged behind standard English in implementing the use of the WH system into written and spoken language.” These observations are supported by numerous reports of a propensity for that relative markers in Scots (Aitken, 1979:105; Macaulay, 1991; Murray, 1873:197; Wilson, 1915:91) and Irish (Garvin, 1977:110; Harris, 1984:131; Miller & Brown, 1982:16–17). The results reported here, which come from Scotland, Ireland, and Northwest England, may be interpreted as simply following along with the expected pattern for these climes – propensity for that relatives and lack of WH relatives.

However, one of the most consistent observations about spoken British English, more generally, as well as nonstandard dialects, in particular, is the same propensity for that and concomitant lack of WH words (Harris, 1967; Tagliamonte, 2002b; Tottie, 1997; Tottie & Harvie, 2000; Van den Eynden, 1993). In contrast, research that reports high rates of WH words is typically based on much more standard varieties of English (Guy & Bayley, 1995:153; Kikai et al., 1987; Quirk, 1957; Sanders, 2002; Tottie, 1995).

To set these overall frequencies within the context of historical change, sociocultural circumstance, and stylistic considerations, we now return to consider the trajectory of development of the WH forms in English and how and why they entered the system in the first place.

A change from outside?

The trajectory of change for the use of who indicates that this marker penetrated the grammar of English as a change from above. Mustanoja (1960:199–200) suggested that WH was adopted from the Romance acrolects of French and Latin. Romaine (1982a:212) agreed, as “the entrance of WH by way of the most complex styles of writing and syntactic positions does suggest a pattern of external prestige borrowing or ‘change from above’ in Labov's terms.” Moreover, there is a great deal of consensus that the WH forms came into the relativization system through formal registers (Dekeyser, 1984:76; Montgomery, 1989; Peters, 1996:125–126; Romaine, 1982b:234; Tottie & Rey, 1997:465). Dekeyser (1986:293) suggested that even once established, the WH forms remained high-register variants. Barber (1997:214), for example, found little use of who in Restoration comedy, leading him to conclude that “the growing dominance of who is typical of a formal literary style. In informal style, and probably in speech, that remained the dominant relative pronoun.” Thus, all indications are that the rise of the WH forms for relativization in British English was a “learned change” (Nevalainen & Raumolin-Brunberg, 2002:110).

The interpretation of WH relative markers as a change from above is supported by quantitative data from the historical record. Analysis of early English correspondence reveals sociolinguistic conditioning from the very beginning (Nevalainen & Raumolin-Brunberg, 2002:120). Use of who begins in highly formulaic contexts, such as closing formulae for letters or reference to deities (Nevalainen & Raumolin-Brunberg, 2002:115–116). Hope (1994:35) offered a sociolinguistic interpretation for the lower rates of who between Shakespeare and the later playwright Fletcher: “although ‘who’ is a rising form … Fletcher avoids it, probably because of it's formal connotations.” Even later, in 17th century American English, there is a marked difference in use between who and that in texts: the ratio between who and that is 12:33 in formal texts and 1:17 in the informal ones (Rissanen, 1984:429). Rissanen concluded that “who is more popular in [formal] texts than in the ones representing a more colloquial style” (Rissanen, 1984:429).

If the introduction of WH was a change from above, it is not surprising that diachronic studies also report “a clear lag between the more literary texts and spoken usage” (Ball, 1996:248). Thus, the question becomes, has this situation changed in the intervening centuries? Tottie's (1995, 1997) extensive research on contemporary varieties of English showed a heavy correlation with education level and situational context, and Macaulay's (1991:65) study of spoken Scots reported an extreme difference between middle-class and working-class speakers.

Table 7 attempts to synthesize the available data in the literature by comparing the reported use of who in subject relative clauses across a number of data sets, from which this information can be recouped from the presentation of data.Virtually all the data representing older, peripheral (conservative) varieties and speakers have the very same tendencies – a low frequency of WH. This is especially revealing given the southern geographic locations of Suffolk, Dorset, Devon, and Somerset. Thus, put in context with available dialect data, the relative marker patterns in Lowland Scotland, Northern Ireland, and Northwest England may not simply be the result of Scots or Irish or northern varieties in Britain having a propensity for that. The lack of WH may instead be a reflection of the more general dichotomy between such linguistics features, as rural versus urban; spoken versus written; standard versus non-standard language; and core versus periphery (Poplack & Tagliamonte, 2001).

Overall frequency of who in subject relatives across British studies

Thus, we suggest that the trajectory of linguistic change in the (written) historical record (e.g., Ball, 1996) made it appear that the WH forms were the favored forms in English and would eventually take over as the dominant relative markers. However, these results show nonstandard dialects – and these northern Scots, Irish, and English ones in particular – steadfastly resisting that trajectory. The WH forms have not yet penetrated their vernaculars; and may never do so.

Notwithstanding the lack of WH forms however, there is tremendous variability between zero and that (see Table 5). The next step then is to explain the alternation between that and zero. What is the status of these forms in the grammar of these northern British dialects?

Pronouns or complementizers?

It is important to keep in mind that the variants of WH, that and zero, are thought to have different categorical status in the syntax of English (e.g., Denison, 1998:276). WH relative markers are commonly agreed to be pronouns (Cofer, 1975:14–15; Huddleston, 1984, 1988:156; Quirk, 1957:98–99; Quirk et al., 1985:366; Radford, 1988:480; Romaine, 1982a:234; Sigley, 1997). On the other hand, that and zero are said to be more like complementizers (Dekeyser, 1984, 1988; Huddleston, 1971:229, 1988:156; Van den Eynden, 1993:65–66).

WH relative markers are considered pronoun-like because they are marked for (1) case, who and whom, and (2) humanness, who and which.¹⁹

From these differing forms and distributional patterns, we might hypothesize that, unlike the WH forms, that and zero are governed by the same set of constraints, and are merely “optional” from a syntactic viewpoint. However, if they are the same syntactically, then what is the reason for the robust variability between them? These data present an opportunity to uncover the underlying constraints governing the choice of one over the other. Thus, for the remainder of the article we will focus on the productive variability between that and zero.

Subject relatives

Perhaps the most striking difference between that and zero is their frequencies in subject and nonsubject relatives. Although use of the zero form is acceptable in Standard English when the relative marker's function is an object (Denison, 1998:280; Quirk et al., 1985:420), when it is a subject, zero is acceptable only in informal or colloquial styles (Quirk et al., 1985:420).

Indeed, subject zero relatives²⁰

Sentence structure

Perhaps the most widely attested factor favoring the zero relative is sentence structure. Existential constructions, as in (28), clefts, as in (29), and possessives have and got constructions, as in (30), are well-known contexts propitious to the zero form.

Because the frequency of the zero variant in these contexts differs so much from all other constructions, as well as in discourse function, their syntactic status as relatives is controversial (Ball, 1996:235; Quirk et al., 1985:1407).

With clefts, Quirk et al. (1985:416) stated that WH markers are rare in this context compared to that and zero. From a discourse level perspective, existentials and clefts also differ from prototypical restrictive relative clauses in that they introduce new information in the discourse (Henry, 1995:125).

However, Ball's (1994:198) quantitative analysis of clefts in the historical record led her to the conclusion that “they are not syntactically distinct” from other types of relatives. In a more recent analysis (Ball 1996:236) however, she excluded them, given the “slightly lower observed frequencies for WH in clefts over time.”

Here, we include all these constructions in order to view their different distributional patterns in the data compared to other constructions, as well as to compare these patterns with those found in earlier dialect studies.

Length of relative clause

Cofer (1975:31) suggested that an overt relative marker (WH or that) is used when the relative clause is long. He claims that this clarifies the structure and aids comprehension of what follows. In other words, processing constraints operate. Following well-known procedures in psycholinguistics, we coded for number of words in the following relative clause, as in (31–32)(Clark & Clark, 1977:58). These were coded based on word count (e.g., one word, two words, etc. up to ten words and over). However, distributional analysis revealed that the dividing line in usage patterns of that and zero was between what we term short clauses (1–3 words), as in (31), and longer clauses (4+ words), as in (32).

However, length does not directly map onto complexity (Chomsky, 1975). To test for both length and complexity, we coded for whether the relative clause was simple, that is, a verb and its nonclausal arguments, as in (33); or complex, that is, containing a clausal argument, as in (34).

Crosstabulation of our data revealed that these two factors were highly interactive. Predictably, most of the long clauses were complex. However, short clauses, which could be either simple or complex, were split dramatically with respect to the appearance of an overt relative marker, as we shall see. These interactions made analysis with the two factor groups together unreliable.²¹

Type of antecedent

Another important constraint relating to the appearance of relative markers is the type of antecedent. Huddleston (1971) reported that indefinite antecedents favor the overt relativizers WH and that, whereas definites as well as pronominal antecedents more generally are said to favor the zero form (Guy & Bayley, 1995:151; Tottie & Harvie, 2000:214). On the contrary, Tottie and Harvie (2000:209) found that indefinites favored zero in subject position, and Kautzsch (2002:178) reported a rather mixed pattern of use across his five data sets in this environment.²³

However, we discovered that definite pronouns, as in (35), rarely figure as antecedents for relative clauses in these data.²⁴

Adjacency

Adjacency is another factor that has been mentioned in the literature as constraining the use of a relative marker. It is claimed that an overt marker is favored in cases where there is intervening material, as this makes the boundary between the matrix and subordinate clause clearer (Cofer, 1975; Guy & Bayley, 1995:155). To capture this distinction, we distinguished adjacent antecedents, as in (39a), from those that were nonadjacent, as in (39b).

Sentence type

The strongest of the internal constraints is sentence type. Three types of constructions are set apart from the others – existentials, possessives, and clefts. In general these favor the zero form, although there are certain differences across communities.

In subject position the use of the zero relative in these constructions is widely attested and acceptable as informal or colloquial (Quirk et al., 1985:1407). In these nonstandard dialects use of zero hovers close to or near 20%. However, it is important to keep in mind that the occurrence of zero is virtually restricted to these specific constructions – existentials, clefts, and possessives.²⁷

In nonsubject position, where zero is a standard variant, sentence structure was also shown to condition the use of zero, but the constraint ranking for sentence type is somewhat different. First, possessives consistently disfavor zero, unlike in subject position where they consistently favor zero. Existentials and possessives are relatively rare. Second, clefts favor zero in Lowland Scotland and Northwest England, and the zero form is favored in existential constructions for Lowland Scotland and Northwest England only. Finally, sentence type is not selected as statistically significant in Northern Ireland. Such differences hint at certain cross-variety differences in the use of zero in the various syntactic constructions and provide additional evidence that they may well have differential status in the syntax of different dialects.

Although it is difficult to provide an explanation for such low-level differences here, such cross-variety anomalies pinpoint areas of grammar that may be worth pursuing from a dialect-syntax perspective. For example, Henry (1995:135) raised the possibility that, “different phrasal categories may be a dimension that differentiates dialects.” Such differences may be the case here.²⁸

As far as the historical record is concerned, there is little evidence for the particulars of nonsubject use of zero aside from their confinement to focal constructions (Erdmann, 1980:149). Thus, whether the observed patterns – the favoring effect of clefts (Lowland Scotland, Northwest England) and the favoring effects of existentials (all communities) – is a reflex of older patterns of use cannot be established. However, the general result here is that as far as sentence structure is concerned, Northern Ireland appears to be differentiated from Northwest England and Lowland Scotland. The Northern Ireland results, where there is little differentiation in sentence structure (and thus nonsignificant, see Table 9) are reminiscent of Ball's (1994) results for historical data, where clefts were found not to be distinct from other constructions. Thus, it may be that Northern Ireland is preserving this older pattern.²⁹

Complexity/length of relative clause

What can explain the effect of the combined complexity/length of the relative clause, which is consistent across subject and nonsubject function and parallel in constraint ranking in every community? Recall Cofer's (1975:31) claims that an overt relative marker (in this case that) helps to clarify the structure and meaning of a complex relative clause. Moreover, “long complex elements put an extra burden on the parser” (Wasow, 1997:94). In our data, short simple clauses did indeed favor zero, just as this hypothesis would predict. Indeed, this tendency is very strong. However, complex, long clauses did not show a strong tendency for an overt relative marker, and this is found regardless of community or grammatical context. Instead, they patterned in an intermediate way. Thus, contrary to expectation, there is no definitive contrast between simple and complex clauses with respect to the occurrence of an overt relative marker. Instead, clause length explains more of the variation. Indeed, the consistency of this effect lends support to an explanation of the variation based on surface-level processing.

Recall that in the theoretical literature that is considered a complementizer. If so, then the constraints we have found here for relative marker that may be the same as those that would be expected for the appearance of complementizer that. A comparison of relative and complement clauses based on the same set of constraints may provide empirical evidence for this claim (Tagliamonte & Smith, to appear).

A note on cross-variety differences

What perspective have we gained from the comparison across three varieties in three distinct regions of Britain? The overwhelming observation is the number of similarities. The parallels are not only with respect to the frequency of that, but more importantly in the ranking of factors constraining its use – length, type of antecedent, complexity, and to varying degrees, sentence structure for subject relatives. Thus, instead of “no vernacular norm” in the use of relative marker that, its use is highly structured, and there are actually considerable similarities across these dialects.

This raises a fundamental question: How far do the similarities between the communities from these geographically delimited areas match those of other varieties within Britain and elsewhere in the world? In the next section, we attempt to answer this question.

Comparison to other varieties?

First, the overall findings that that predominates as the relative marker in these northern British dialects provide some insight as to why there is an overwhelming tendency for use of relative that in North America (Tottie & Rey, 1997). It may simply be the product of the fact that that was the preferred and more prevelant form in the founder populations.³⁰

Yet despite the overwhelming presence of that in both contemporary and historical reports on British dialects, as well as in North American dialects, there is also the relative marker what to consider. According to the SED, it survived across quite a wide range of southern British dialects (Orton & Halliday, 1963). Some of these varieties also figured in the migrations into North America (Kytö, 1993; Le Page & Tabouret-Keller, 1985; Wright, 2004). So, why did what not get selected into colonial varieties of English in North America? Indeed, recent research shows that what is an extremely common variant in two independent studies of Early African American Vernacular English. It occurs 50% of the time in both subject and nonsubject contexts for Samaná English (Tottie & Harvie, 2000:213), and at rates as high as 51% in the early African American corpora studied by Kautzsch (2002:162). Why relative what would be selected in these varieties and not in other North American dialects studied so far is an enticing question for future cross-variety comparison.

For now, however, we shall restrict our comparisons to the shared retentions across dialects (the two relative markers that and zero) and the constraints that operate on them. Unfortunately, there is a paucity of information on the ranking of constraints for relative markers in nonstandard dialects in North America, which prevents detailed comparison. However, two internal constraints can be extrapolated from the claims and observations in the existing literature. Martin and Wolfram (1998:32) reported that zero subject relatives in subject function in African American Vernacular English “tend to be favoured with copula structures” and are “particularly prominent” in existentials. Further, existentials are said to have the highest frequency of zero relatives. The same patterning can be observed in the constraint ranking across these northern British dialects. Both constructions consistently favor zero in subject function across all varieties under investigation at frequencies that are fairly high. Moreover existentials overwhelmingly outrank clefts in subject/function (see Table 7).

Interestingly, the two sources of comment on the use of subject zero relatives provide dramatically different explanations for the presence of this feature. Martin and Wolfram (1998:32) argued that this patterning represents an area of “notable difference between AAVE and other English vernaculars.” On the other hand, Henry (1995) argued that the availability of the same constructions in Belfast English is a result of the “strongly topic-prominent nature of that dialect.” The same favoring effect for existentials and clefts is reported for Somerset (Ihalainen, 1980:189) and Dorset (Van den Eynden Morpeth & Hogeschool, 2002).³¹

On the other hand, Tottie and Rey (1997) found higher rates of zero in constructions other than existentials and clefts in Early African American Vernacular English. This feature is not consistent with the British dialect data we have considered here, and thus may be an independent development in this particular variety.

Another factor, noted in North American research at least, is the favoring effect of indefinite antecedents. Tottie and Harvie (2000) found that indefinites tended to favor the zero relative in subject relatives in Early African American Vernacular English. Although not selected as significant in our data (see Table 7), the same trend is visible. In every variety, there is a higher propensity for zero relatives with indefinites. Thus, there appears to be no consistent cross-variety pattern for this factor. However, in other corpora, such as those studied by Kautzsch (2002), there is no consistent pattern across varieties. What would explain such discrepancies?

Cross-tabulation of our own data revealed interaction between existential constructions and indefinite antecedents. As Table 10 shows, the vast majority of existential constructions have indefinite antecedents, as in (40).

Distribution of antecedent types in existential constructions by community

This means that the divergent results for indefinite antecedents across varieties may simply be an artefact of the number of existential constructions in the particular data set under consideration, and further, whether or not the dialect in question has a tendency towards zero relatives in those contexts.

In sum, there are at least two factors that constrain that and zero relative markers – length/complexity and syntactic construction. Which additional varieties share these constraints and whether in fact there are any other constraints that typify vernacular dialects are important areas for further exploration. Thus, for example, length effects might be expected to be universal, as these are dependent on cognitive processing constraints, which are presumably shared by all speakers (Fodor, 1998). On the other hand, tendencies of that or zero for clefts and/or existentials may be dialect specific. If so, their ranking vis-à-vis other constructions for relative marker use may prove to be valuable diagnostics of origins and/or interdialectical relationships. It also remains for future research to compare and contrast the internal details of the constraints found here with varieties in North America to assess which are universal and which may be traced back to dialect-specific roots.