BACKGROUND
Example (1) shows two excerpts from an interview with a speaker of
early New Zealand English. Mr. R. was born in New Zealand in 1898. Both
excerpts show an example of singular concord with the verb be.
An example such as that in (1a) would be unusual in contemporary New
Zealand English, whereas that in (1b) would be unremarkable. Such
examples indicate that patterns of subject–verb agreement with
the verb be may have changed over the history of New
Zealand—at least in nonexistential environments such as (1a).
This article traces the use of subject–verb concord with the
verb be through the history of New Zealand English. Singular
agreement with plural NP subjects is examined, in both existential and
nonexistential constructions. We document changes that have occurred
with existentials and nonexistentials, and investigate the degree to
which these changes are independent of one another.
The verb be is prone to variation in many dialects of
English, and the inherent irregularity of the present and past tense
paradigms of the English verb be is thus of vital interest to
variationist linguistics. The five variants (am, is,
are for present, was and were for past
tense) give this verb iconic status as they combine in a verbal
paradigm so irregular that, in the words of Wolfram, Hazen, and
Schilling-Estes (1999:75), it is
“without parallel in the current configuration of
subject–verb concord.”
English verbal morphology is pervasively regular, and distinct tense
variants are minority forms and linguistically marked. Marked variants
are often subject to pressures toward analogy (particularly in
vernacular varieties), that is, they frequently display tendencies to
bring irregular person-number concord in line with the vast majority of
verbal paradigms that display no such agreement. Accordingly, there
have probably always been trends toward paradigmatic regularization,
and most historical and contemporary forms of English are subject to
alternation of variants. For instance, a number of historical studies
attest to the diachronic dimension of be irregularity; Quirk
and Wrenn (1960) documented alternation among
distinct patterns in Old English (particularly in existentials), and
Traugott (1972), Visser (1970), Jespersen (1961),
and Denison (1998) trace its development in
the Middle and Early Modern periods of English.
The historical evidence is complemented by a plethora of synchronic
studies on be variation in varieties of English around the
world. There is a particularly sizeable literature on
was/were alternation (or past be
regularization), which has been investigated in the British Isles
(Britain 2002; Cheshire,
1982; Ojanen, 1982; Tagliamonte, 1998), the United States (Labov, 1972; Labov, Cohen,
Robins, & Lewis, 1968; Feagin,
1979; Hazen, 1994, 2000a, 2000b; Mallinson & Wolfram, 2002; Wolfram & Christian, 1976), the South Atlantic
(Britain & Sudbury, 2002; Schreier, 2002b, 2003), Australia
(Eisikovits, 1991), Canada (Meechan & Foley, 1994), and the Caribbean
(Tagliamonte & Smith, 1998). Whereas most
of these quantitative studies looked into the patterning of variation
and its internal and extralinguistic correlates, research has also
focused on the delimitation of genetic relationships and (potential)
founder effects (Tagliamonte & Smith,
2000) or on contact-induced regularization mechanisms in
new-dialect formation (Schreier, 2002b).
A few general findings emerge from the literature on
subject–verb agreement with be. First, the present
be paradigm is more robust to levelling than the past
be paradigm. Person-number concord involving am,
are, and is by and large adheres to the agreement
rules of Standard English, whereas extension of the pivot form
was (i.e., forms such as we was) is so widespread
that it constitutes a “vernacular root” of English. This
point is made by Chambers (1995:242), who
identified we/you/they was as one of the
“variables [that] appear to be primitives of vernacular
dialects in the sense that they recur ubiquitously all over the
world.” Forms like we was are found in many varieties of
English (Cheshire, Edwards, & Whittle, 1989, 1993), whereas
I/we is are infrequently reported and geographically
restricted to areas such as northern England (Linguistic Atlas of
England; Orton, Sanderson, & Widdowson,
1978), the Southern United States (e.g., rural Mississippi;
Burkette, personal communication, 2002), or the South Atlantic Ocean
(Schreier, 2002a; Schulenburg, forthcoming; Wilson, 1997). Accordingly, there is a particularly
extensive literature on past be realignment, and a few trends
emerge. One important finding is that, even though extension of
was is by far the most common process, multiple pivot forms
may be selected during regularization processes. Alternation between
variants is thus not necessarily a unilateral process. Alternative
regularization trends may occur in one and the same variety, and some
studies report considerable variability in terms of directionality. The
most usual pattern is for was to feature in contexts of
standard were (Chambers' 1995
“vernacular root”), but alternative processes include the
extension of were to contexts of was (e.g., I
were there last night; Cheshire,
1982; Tagliamonte, 1998) or
weren't to the context of wasn't (e.g.,
I weren't planning to come; Britain, 2002). One important implication is that
varieties may develop highly intricate patterns, namely the
simultaneous extension of was to positive contexts and
weren't to negative ones (Britain,
2002; Schilling-Estes & Wolfram,
1994). Such two-fold leveling processes result in a realignment
of the two paradigms, contributing toward a “remorphologization
of the was and were allomorphs of past be
along positive/negative, rather than person-number, lines”
(Schilling-Estes & Wolfram, 1994:280). As
a result, both was and weren't allomorphs may
occur in all subject contexts, with both numbers and in all persons.
This pattern is also found in the contemporary English Fens English,
where Britain (2002:17) reported that
“Young people […] increasingly use was
in affirmative clauses (irrespective of person/number) and almost
exclusively use weren't in negative contexts.”
Several morphemes may thus undergo parallel extension and reanalysis,
and paradigm regularization does not always operate as a unilateral
process.
Another finding to emerge from the literature is that alternation
between be variants is subject to robust linguistic
constraints, and that individual contexts exhibit differential
tendencies to cooccur with extended forms. The type of subject plays an
important role and, as Tagliamonte (1998)
pointed out, modern variationist studies often focus on third-person
plural contexts, as this environment offers particularly important
insights into linguistic constraints on variation. Third-person plural
environments can cooccur with a personal pronoun (they were),
a NP (the cows were), or with an existential subject
(there were cows). These three environments have different
agreement patterns, with existentials having the highest levels of
was agreement (there was cows), and personal pronouns
having the lowest (we/they was, with forms like we
is being very infrequent). This finding is replicated in
practically all studies on subject–verb agreement involving
finite forms of be (Cheshire, 1982;
Eisikovits, 1991; Feagin,
1979; Hazen, 2000a; Wolfram & Christian, 1976). Of the personal
plural pronouns, there is evidence that they is the least
likely to occur with was (Eisikovits,
1991; Feagin, 1979). Second-person
singulars tend to have higher levels of was agreement than
first- and third-person plural contexts (Eisikovits,
1991; Feagin, 1979; Smith & Tagliamonte, 1998; Tagliamonte, 1998). This most likely represents a
reflection of earlier forms of English, which had particularly
extensive was/were alternation in this context.
As Forsstrœm (1948:22) noted,
second-person singulars were subject to considerable regional variation
in Middle English. There was a sharp division between the south of
England, with predominant usage of were, and northern
varieties, which historically used was with this person.
Plural NP subjects, on the other hand, tend to exhibit higher levels
of was agreement than personal pronouns, and studies such as
Feagin (1979), Christian, Wolfram, and Dube
(1988), and Hazen (2000b) reported that the type of plural NP (simple,
collective, or conjoined) may have an effect on agreement patterns,
with collectives and conjoined NPs showing higher levels of nonstandard
concord than simple NPs.
In some dialects, the type of subject has categorical effects. For
example, the Northern Subject rule has operated in the northern
dialects of Britain from the 13th century (Montgomery, 1994; Murray,
1873). The Northern Subject rule allows verbal -s
endings (including is/was forms of be) after full
NPs, but not after pronouns. However, are/were is also
used with full NPs in the case when a clause intervenes between the
subject and verb. It is thus a combination of a type of subject
constraint and a proximity condition.
Of particular relevance is the absolute and relative positioning of
verb and subject phrases (Britain & Sudbury,
2002; Meechan & Foley, 1994).
Was agreement is much more likely to operate when the verb
precedes subject NPs (for instance in existential there V NP
constructions) than when the NP occurs in canonical pre-V position.
Existential plural constructions have “by far the strongest
effect on the use of was in contexts of standard
were” (Tagliamonte & Smith,
2000:157), and this finding is reproduced in all varieties that
have been subject to linguistic scrutiny (Christian
et al., 1988; Eisikovits, 1991; Hazen, 2000b; Schilling-Estes
& Wolfram, 1994; Schreier, 2002b).
There is also some evidence for an effect of the proximity of the verb
and the subject, with intervening linguistic material facilitating
nonstandard concord (see, e.g., Britain &
Sudbury, 2002; Tagliamonte, 1998).
Even though existentials are usually discussed in conjunction with
other linguistic environments, some recent studies have investigated
this context separately, with different intentions. Meechan and Foley
(1994:63), assuming that “the unusual
concord variation pattern was a result of structural differences
associated with the restriction on the type of determiners preceding
the postverbal NP”, embedded their analysis in a detailed
syntactic and phonological framework. They included factors such as the
assignment of case and the control of agreement, and considered the
implications of there's/there are alternation for
syntactic theory and socially or language-internally conditioned
effects on variation (Meechan & Foley, 1994:65 ff.; see also, Cheshire,
1999). Britain and Sudbury (2002), on
the other hand, investigated existential constructions in terms of
their potential for parallel language development in two transplanted
forms of English (Falkland Island English, FIE, and New Zealand
English, NZE): “the universality of variability in (there's)
[…] enables us to assess whether the diffusion of
(there's) variability across the anglophone world has proceeded in
the same way in each speech community” (Britain & Sudbury, 2002:213). They compare a
variety of linguistic constraints on agreement with existentials
(tense, determiner and quantifier types, distance between VP and NP,
etc.) and document remarkable similarities in the two varieties, which
they discuss with reference to parallel independent language change and
the controversial concept of language “drift” (Sapir, 1921; see also, Hickey,
2003; Trudgill, Gordon, Lewis, &
Maclagan, 2000).
Subject–verb agreement with be is consequently one of
the most detailed and widely researched variables in English; analysis
of this feature has contributed toward the study of language change,
variation analysis, dialect transplantation and contact, and genetic
linguistics. The present study adds to the literature, focussing on the
evolution of concord patterns in a dialect contact scenario. We address
questions such as: What trajectory does subject–verb agreement
(of present and past tense paradigms) follow in this (post)colonial
locale that involves the transplantation of several inputs? Are there
parallels and similarities with other varieties, or does concord follow
its own path, deviating from dialect to dialect? Is the development of
subject–verb agreement patterns in existential constructions
linked to patterns of usage in nonexistentials? With these aims, we
trace the historical development of subject–verb agreement for
the verb be in New Zealand English.
DATA AND METHODS
Archives
The data analyzed come from three archives, all of them held by the
Department of Linguistics at the University of Canterbury,
Christchurch. One of them is the Mobile Unit (MU) archive, the basis of
the Origins of New Zealand English (ONZE) project (Gordon, Campbell, Hay, Maclagan, Sudbury, & Trudgill,
2004). The recordings in the MU database were conducted in
1946–1948 by members of the Mobile Disc Recording Unit of Radio
New Zealand, whose principal aim was the collection of personal
reminiscences of the earliest phases of British involvement in New
Zealand (Lewis, 1996; Woods, 1997). This archive contains recordings of
more than 250 people born in New Zealand between 1853 and the early
1900s, and is the only collection of spoken data from 19th-century NZE
available to date. Most of the people interviewed were first or second
generation New Zealanders, that is, children of the first European
colonizers of New Zealand (primarily from England, Scotland, Ireland,
and the Australian colonies). The MU archive has been subject to
intensive research and has yielded a wealth of information on contact
dialectology in general (Trudgill, 1998;
Trudgill et al., 2000) and the formation of
New Zealand in English in particular (Gordon et al.,
2004; Sudbury & Hay, 2002; Trudgill, Gordon, & Lewis, 1998). The other two
archives, which are subject to extensive current research, contain
recordings of New Zealanders born in the 20th century. The intermediate
archive (IA) contains recordings of about 150 New Zealanders born
between 1896 and the early 1930s. Most of these data were collected for
oral history projects, which may be of importance because they are
stylistically less formal than the MU recordings (further discussion
later). The third archive, the Canterbury Corpus (CC), contains some
370 speakers. This is an archive created by the University of
Canterbury New Zealand English class, which each year interviews
individuals fitting into a prespecified sample, stratified by age,
gender, and social class. Together, the three corpora contain over 700
speakers born between 1853 and 1980, and span the entire formation
period of NZE. This article analyzes subject–verb agreement in a
subset of 146 of these speakers.
Extraction and coding procedures
This article has as its origin a class research project assigned to
Ling 203 – Sociolinguistics at the University of Canterbury, New
Zealand, taught in 2002. In a preliminary stage, each of the 47
participating students was assigned several transcripts (taken from the
three corpora described) and took responsibility for coding these for
subject–verb agreement.
Students were instructed to extract and code in the transcripts all
examples of be with a plural subject, excluding noncount cases
such as nonexistential inverted examples (how are the
children?). Examples involving mass nouns (the fish
weren't present) were also not included. The data analyzed
here were thus initially extracted by students and subsequently
double-checked by both authors. Examples of be with singular
subjects were not analyzed. This is because preliminary analysis of the
data revealed that instances of nonstandard
are/were were extremely rare in the data set. The
quantified analysis was therefore restricted to nonstandard
is/was, which preliminary analysis revealed
occurred with relatively high frequency.
Note that the analysis was based on orthographic transcripts of the
recordings. These transcripts include detail about hesitation and
overlap, and have been double-checked by multiple research
assistants.
Some attempt was made to analyze a balanced stratified sample,
although this was partially limited by practical considerations. The
two most important constraints are that the earliest MU corpus contains
more male speakers than female speakers, and that not all speakers are
stratified along socioeconomic criteria. The most recent corpus makes a
distinction between “professional” and
“nonprofessional” informants, a distinction which is made
both in terms of level of education and type of occupation (see
Maclagan, Gordon, & Lewis, 1999, for
details). Class-based variables are therefore available for the
later-born speakers only. Table 1 shows the
social distribution of the 146 speakers analyzed. Entire transcripts
were analyzed for all speakers. The recordings differ quite
substantially in length, and so the amount of data available differed
across speakers. The amount of speech transcribed ranges from 10
minutes to over an hour. Note, also, that there is a slight overlap in
birthdates between the MU and the IA, and between the IA and the CC
speakers.
Social distribution of the 146 speakers analyzed
Our analysis considered a number of social and linguistic criteria.
Social factors considered here include age and gender (and, for
speakers for which such information was available, social class).
Consequently, all tokens were coded for date of birth and gender of the
individual speakers. Lastly, speakers from the CC corpus were coded for
social status (professional vs. nonprofessional). We also analyzed a
number of language-internal factors, namely:
- Tense (past, present)
- Polarity (positive, negative)
- Modifier type [bare (i.e., no modifier), definite determiner,
number, negative, adjective, quantifier, ‘a’ quantifier
(e.g., a lot of, a few)].
- Distance between noun and verb (measured in the number of
interpositioned words: 0, 1, 2, 3+, or ‘a’ in cases with no
overt subject)
- Subject type (regular plural, irregular plural, third-person
pronoun, first-person pronoun, conjoined NPs)
- Contractedness of the verb (contracted, notcontracted).
For each data set, the discussion focusses on those factors which had
a significant effect on the data. Some factors (e.g., polarity) did not
prove significant in either data set, and will not be discussed
further.
RESULTS
The final data set contains 1028 tokens of existentials, and 4364
examples of nonexistentials (total = 5392). We present the results of
the existentials and the nonexistentials separately from one another,
before considering how the two sets of results relate. We begin with
the existential forms.
Existentials
Figure 1 shows the pattern of singular
concord with existentials over the period studied. Note that the
pattern is distinctly nonmonotonic—use of the singular appeared
to decrease during the late 19th century, but then the trajectory of
change appears to have reversed, with a relative increase of the
singular. This is an unusual trajectory, and certainly does not follow
the S-shaped curve which is typical of language change (see, e.g.,
Bailey, 1973; Kroch,
1989; Weinreich, Labov, & Herzog,
1968). We will return later to the question of how or why such a
reversal might have happened. Note, too, that there is a clearly
gendered pattern during the late 19th century, with female speakers
consistently producing higher rates of singular concord. This pattern
is not present among the later speakers. Table
2 shows the breakdown of the data underlying Figure 1.
Effect of birth date and gender (existentials)
Effect of birth date and gender (existentials).
Although we don't have social information available about the
earlier speakers, it is available for those speakers belonging to the
Canterbury Corpus. Figure 2 and Table 3 show the pattern relating to social class
and use of singular concord. Professional speakers show a marked
reduction in singular concord relative to nonprofessional speakers.
Note that this holds equally for male and female speakers.
Rate of singular concord with existentials: Effect of gender and
professional status (Canterbury Corpus only)
Professional status (Canterbury Corpus only). This figure is broken
down by male and female to show the results for each gender, but note
that there is no significant interaction between professional status
and gender.
There are three linguistic factors that also play an important role:
modifier type, tense, and contractedness. The effect of modifier type
is shown in Figure 3 (and Table 4). Following Britain and Sudbury (2002), we coded the following seven categories:
Preference was given to the initial quantifier. For example, “a
lot of the cats,” would be coded as an “a”
quantifier, and “many red cats” would be coded as a
quantifier, rather than an adjective. And coding related to linguistic
form only. Thus, although there are many ways to invoke definiteness
without the presence of a definite article, the code
“definite” indicates the presence of such an article.
The effect of the type of modifier on singular concord with
existentials.
Rate of singular concord in existentials: Effect of modifier
type
While adjectival, bare NPs and quantifiers showed reduced rates of
singular concord, comparatively high rates were returned with
definites, negatives, numbers, and quantifiers containing
“a” (e.g., a heap of people).1
Note that the varbrul model of the data (Table 10)
indicates that a bare NP leads to marginally more singular concord than
a quantifier, contrary to the raw data in Figure 3.
These
results are largely in line with previous results investigating
modifier type. Meechan and Foley (
1994)
reported a hierarchy of “no > number > weak determiner
/strong determiner,” and Tagliamonte (
1998) found “no > partitives >
definites > numbers > quantifiers > bare NPs” (see
further discussion later).
Finally, both tense and contractedness are important determiners of
rate of singular concord. There is an important interaction between
these two factors, as can be seen in Figure 4
(Table 5). Examples (2a–d) illustrate
contracted and noncontracted forms of the verb with singular concord,
in decreasing order of likelihood. Note that the surface form of
singular concord with contracted verbs in past tense is the same as
present tense—context was used to distinguish which was intended.
As much of the data involves speakers reminiscing about events in the
past, context usually contained clear tense information.
Effect of tense and contractedness (existentials).
Rate of singular concord with existentials: Effect of tense and
contractedness
Both contractedness and tense play an important role, and there is a
clear interaction between them. Figure 4
indicates that there is a strong interaction. While past tense
increases the likelihood of singular concord for noncontracted forms,
it decreases the likelihood for contracted forms. We can perhaps
account for the extremely high rate of singular concord with sentences
such as (2a) by ease of articulation of contracted
“there's” forms.
“There's” is one syllable, and so provides a
strong articulatory advantage over “there is,”
whereas “there're” contains the same number
of syllables as “there are,” and so does not
provide a particularly strong advantage. Of course, many speakers will
contract “there're” down to a single
syllable. We can speculate that this may also be dispreferred,
(relative to “there's”), because of the
resulting ambiguity between “there” and
“there're”
Table 6 shows the varbrul model of
the entire data set. Note that the interaction between gender and birth
date has been dealt with by combining the two into a single factor
group. Three age groups have been defined: those born before 1890,
those born from 1890–1949, and those born after 1950. This
grouping reflects the patterns revealed in Figure
1, where those born before 1890 and after 1950 appeared to have
the highest rates of singular concord. The varbrul model
reveals that the reversal in trajectory is robust for the female
speakers, but not statistically reliable for the male
speakers—the varbrul weights systematically reduce for
the male speakers, with the earliest born speakers using most singular
concord, and the later born speakers using least.
goldvarb results for singular concord in existentials over
entire data set
Tense and contractedness were also combined into a single factor
group, which was retained as significant, and reflects the patterns in
Figure 4. Modifier type was also retained as
significant. Note that the varbrul weights for modifier type
reverse the order of “bare NP” and
“quantifier,” relative to the ordering shown in Figure 3. Other linguistic factors were also
included in the model (as itemized earlier), but these were not
retained as significant.
Although professional status was not included in the model of the
overall data set, a separate model was fit, based on the Canterbury
Corpus only. In this model, professional status was also retained as an
important predictor of the data (professional: varbrul weight
= .334; nonprofessional: varbrul weight = .696).
Before turning to the results for the nonexistentials, it is worth
considering a trend which was not retained by the varbrul
model. In their analysis of (there's) in New Zealand
English and Falkland Islands English, Britain and Sudbury (2002) reported a significant effect of
distance—the larger the number of lexical items positioned
between the noun and the verb, the more likely singular concord was to
operate in their data. Tagliamonte and Smith (2000) also reported a similar effect for York. Our
results concerning distance, while not significant, do show the same
pattern (Figure 5).
Distance between verb and noun. This shows the effect of how many
words intervene between the verb and the noun. The result shown in this
graph is not statistically significant, given the other factors already
discussed.
Our results indicate a (nonsignificant) trend for singular concord to
increase with the distance between the subject and the verb. We turn
now to the presentation of results concerning nonexistentials.
Nonexistentials
A total of 4364 tokens of nonexistentials were coded. Our analysis
showed that singular concord with nonexistentials was very rare in both
the Intermediate Archive and in the Canterbury Corpus. Figure 6 breaks down the rate of singular concord
over the entire time period, for men and women. Speakers born in the
20th century are grouped together in a single category for the purposes
of this graph, as there is practically no variation in this period.
From 1900 on, singular agreement with nonexistentials is virtually
nonexistent in NZE. In 19th century NZE, however, we find that singular
concord with personal pronouns (they was) and plural NPs
(the cats was) occurs more frequently, and that men have
markedly higher rates than women. We also see a steady decrease of
singular concord throughout the second half of the 19th century. It all
but disappeared by the turn of the century and is not a characteristic
feature of present-day New Zealand English.
Use of singular concord in nonexistential constructions, by gender
and birth date.
Because the rate of singular agreement in speakers from the later IA
and CC corpora was so low, an investigation of linguistic effects on
concord variation would seem largely futile. We therefore decided to
confine our statistical analysis to the MU speakers, since they were
born in the period where we find the most extensive variation;
accordingly, the analysis presented here concentrates on data from
19th-century NZE only, with a data subset of 1064 tokens. The same
variables were coded here as in the existentials earlier, with one
exception. The type of the modifier was coded for existentials, because
related studies (Meechan & Foley, 1994;
Tagliamonte, 1998) have indicated the
importance of this variable for existentials. Since the literature we
consulted reported no such finding for nonexistentials, we decided not
to code for modifier type for our nonexistential data. It is perhaps
not surprising that the literature should reflect an important role of
the modifier type in the case of the existentials but not the
nonexistentials. As illustrated in (3a), in the case of existentials,
the modifier is positioned near to the verb—between the verb and
the head noun. In nonexistentials (3b), the noun intervenes between the
modifier and the verb, and it is the noun itself that is likely to be
more proximate to the verb.
The role of gender and birth date, and their interaction, can be seen
in Figure 6. For those born before 1900,
singular concord steadily decreased for both men and women. However,
the decrease was more dramatic for the male speakers than the female
speakers. Thus, as with the existential data, there is an interaction
between gender and birth date, which needs to be taken care of in the
statistical model. The data for the Mobile Unit speakers (i.e., those
speakers who are included in the statistical model) are given in Table 7.
Rate of singular concord in nonexistential constructions by gender
and birth date (Mobile Unit data only)
Figure 7 (and Table
8) indicates the very strong effect of tense on concord
patterns. Singular concord is much more likely to occur with past-tense
reference (the dogs was happy) than in present tense (the
dogs is happy). This is consistent with the large literature on
subject–verb agreement with be, which has consistently
found that the past-tense paradigm is more prone to regularization than
the present-tense paradigm (see Chambers,
1995; Feagin, 1979; Hazen, 2000a; etc.).
Rate of singular concord in nonexistentials: Effect of tense (Mobile
Unit data only)
Effect of tense on singular concord in nonexistentials.
The next effect concerns the type of subject NP. We looked into the
effects of:
The “conjoined” category included cases in which both
singular and plural NPs were combined. However, as there were only 36
conjoined tokens in total, and only three of these exhibited singular
concord, we are not in a position to examine whether the type of
compounding affects the likelihood of singular concord. There were also
three instances where there was no overt subject NP. These cases were
excluded from the statistical analysis.
Two absences from this list are second-person plural pronouns and
collective nouns. As the majority of the recordings involve just two
speakers, the opportunity for the occurrence of second-person plural
pronouns was extremely low. This, combined with the potential ambiguity
between second-person singular and second-person plural, led us to
exclude the second-person plural pronoun from our study.
We also excluded collective nouns, as it has been shown that
collective nouns in NZ English show a large degree of variation in
terms of whether they prefer singular or plural concord. For example,
in Hundt's (1998) study of New Zealand
newspapers, team showed a strong tendency toward singular
concord, police almost never occurred with singular concord,
and family showed considerable variation. As there is a
clearly lexical effect here, and as we were unlikely to collect
sufficient data to study this on a word-by-word basis, we chose to
exclude these from our study.
Table 9 shows the effect of type of subject
NP, and shows that the highest rates of singular concord are found when
the subject NP is not overtly present (many was waiting for their
money). Singular concord decreases with regular plurals,
compounds, and pronouns. Irregular nouns (e.g., children) show
the lowest rates of singular concord in our sample.
Effect of type of subject on rate of singular concord (Mobile Unit
data only)
As outlined earlier, many studies have shown that existentials have
the highest levels of was agreement; personal pronouns, on the
other hand, have the lowest, with lexical items obtaining intermediary
levels (e.g., Cheshire, 1982; Eisikovits, 1991). The results presented here are
largely in line with other findings. Existentials have the highest rate
of singular agreement, and we also note a type of subject NP effect:
regular plurals are more prone to singular agreement than pronouns. The
low rate of occurrence with irregulars is unexpected. We distinguished
between regulars and irregulars because related studies found some
effect here (Britain & Sudbury, 2002;
Meechan & Foley, 1994:76), reporting
higher rates of singular concord if an overt plural marker was absent.
Here we have the opposite result. One possible explanation for this
difference is that Meechan and Foley and Britain and Sudbury included
collective nouns in their studies, whereas our study excluded these
(for reasons discussed earlier). If NPs such as “team” or
“family” were included in our data set, this would no doubt
have dramatically increased the probability of singular concord with
NPs without an overt plural marker. The direction of the difference we
have here is nonetheless surprising and somewhat counterintuitive. It
should also be noted that the number of tokens concerned is not large
(51), and the number of distinct words involved is relatively few.
The results we have outlined: subject type, tense, and the
interaction between gender and birth date were the only factors that
were retained as significant in the varbrul model. This model
is shown in Table 10.2
Note that there were also several tokens with a quantifier
and no overt head noun, but these have been omitted from the
statistical model because of insufficient data.
To model the
interaction between gender and birth date, speakers were divided into
two groups—those born before 1870 and those born 1870 or later.
The
varbrul weights in
Table 10
provide a direct reflection of these results.
goldvarb results for singular concord in nonexistentials
(Mobile Unit data only)
Two effects that were not retained in the varbrul analysis
are also worthy of note. First, Table 11
shows the frequency of singular concord by number of items positioned
between noun and verb. There is some indication of relationship between
frequency of singular concord and number of lexical items between the
subject and the verb. There were only ten tokens with two intervening
lexical items, and so the categories for two and three+ intervening
items have been combined together. Table 11
indicates that tokens with more than one lexical item intervening
between the subject and verb may be more prone to singular concord.
However, the total number of such cases is low, and the factor was not
retained in the varbrul model. So, it is unclear whether more
data would show this generalization to be robust.
Effect of intervening lexical items on concord in nonexistentials
(Mobile Unit data only)
Finally, Table 12 shows the overall results
for contractedness. Contractedness was not considered in the
statistical model, due to one of the cells being completely empty. Our
corpus contains no examples of contracted singular be in
nonexistential forms, and sentences like “the people's very
happy” or “they's coming” are not attested.
Although we are unable to test the level of significance of this
generalization, the complete absence of singular contracted forms in
this context is quite striking. This may partly result from the fact
that these tokens include regular NPs, where singular agreement would
involve the conjunction of a contracted “s” form and an
item that contains plural -s marking. There is some
possibility that such cases occurred, but were not clear from the
context, and were missed by the transcriber (e.g., “the
dogs's happy” could have been transcribed as “the
dog's happy”—if the fact that the noun is plural is
not clear from the context—such cases would then not be included
in our data set, which considers only plural subjects). However,
regular plurals only constitute 24% of the total tokens. The majority
of tokens (68%) involve pronouns, and none of these examples involve
contracted singular forms of be. Also note, however, that the
total number of contracted forms is, in fact, quite low, and it is
likely that a larger data set may have produced some examples involving
singular concord in a contracted form.
Effect of contractedness on concord in nonexistentials (Mobile Unit
data only)
DISCUSSION
The data presented here shed light on important aspects of
subject–verb concord in historical and contemporary NZE. First of
all, our study traces the development of concord patterns in a colonial
dialect of English where this feature has not been subject to previous
analysis. Moreover, given the time-depth of the database, our analysis
offers the rare opportunity to explore the entire evolution of
subject–verb concord in a single variety. Second, our study is of
importance for scrutinizing founder effects and subsequent
accommodation/formation processes in NZE: Exactly what concord
patterns were transplanted to New Zealand, and exactly how did these
patterns develop and interact when the colony formed and a localized
phonological system evolved? Our findings yield insights into the
diachronic dimension of subject–verb agreement, since our data
span more than 120 years in apparent time. How are we to explain the
differential developments of present concord in the two environments
analyzed? Why did singular agreement marginalize in one context
(nonexistentials) and thrive in the other (existentials), despite the
fact that it decreased in both during the 19th century? And, as a last
point, what parallels can we draw between NZE and other varieties? Does
subject–verb agreement align distinctly in these varieties, or do
we find parallels and common characteristics? If so, what insights does
a cross-dialectal comparison yield on the nature of concord in English?
We address these issues in turn.
Founder effects and subsequent developments
Based on the apparent-time principle (Bailey,
Wikle, Tillery, & Sand, 1991), namely that New Zealanders
born in the 1850s continually represent the earliest forms of NZE, and
assuming that our data represent an adequate reflection of the original
input varieties to NZE (see Gordon et al., 2004 for extensive discussion), an examination of
the MU speakers gives us insights into the first stages of this variety
of Southern Hemisphere English. We are therefore in the position to
delimit the concord patterns transplanted and used by the first
generations of New Zealand born colonizers. A first and most immediate
finding is that patterns that are well-established elsewhere are not
found in our data, most notably the extension of was to
positive and the extension of weren't to negative
environments. Leveling to weren't is well-attested in
American English, where it “now seems to be confined to a primary
dialect region along the Mid-Atlantic coastal area that extends from
the Eastern Shore of Maryland and Virginia […] to the
Outer Banks barrier islands and the adjacent coastal region of mainland
North Carolina” (Wolfram & Thomas,
2002:69). We saw that this pattern is attested in the British
Isles as well, most notably in the Fens area in the eastern English
Midlands (Britain, 2002). In contrast, our
analysis revealed that NZE, neither in earlier nor current forms, has
tendencies of realigning be paradigms to express
positive/negative polarity (as documented by Hazen, 2000b; Schilling-Estes
& Wolfram, 1994). Not only is there no evidence of
realignment along polarity lines, but there is no evidence that
polarity plays any role. Polarity was not retained as a significant
factor for either the existentials or the nonexistentials (singular
concord in existentials: positive = 7.77%, negative = 8.3%; in
nonexistentials: positive = 58%; negative = 60%). Thus, while many
other varieties show a clear effect of polarity (see, e.g., Tagliamonte & Smith, 2000), this appears to
play no role in New Zealand English.
We can make the same point for leveling to were, which is
attested in Yorkshire English (Tagliamonte,
1998) or Southwestern English English (Cheshire, 1982). Our preliminary analysis revealed
that forms such as she weren't are practically
nonexistent in our corpus, and this strongly suggests that this pattern
was not brought to New Zealand. Again, New Zealanders born in the
crucial 1850–1880 formation period of NZE have practically no
I/she/the cat were forms, and the most plausible
explanation is that the inputs either had very low levels or did not
have this pattern at all. On the other hand, the most frequent
agreement pattern in our corpus (of course, apart from a standard-like
alternation between was and were) is the extension of
was, and we suggest that this yields insights into input
properties as well. Regularization to was is commonly reported
in Southeastern English and Scottish English (Orton
et al., 1978; Wakelin, 1977, 1986), two varieties that are generally considered
principal contributors to early NZE (Gordon et al.,
2004).
As Figures 1 and 6 indicate, the input properties were modified
considerably, and the crystallization period of NZE witnessed dynamic
leveling processes (at least with regard to the feature investigated in
the present study). The second half of the 19th century was a period of
continuing decrease of singular concord with both existentials and
nonexistentials. Originally transplanted patterns do not remain static
and may undergo quick adaptations and modifications in contact
settings. The trend we note for 19th-century NZE is one of
standardization (i.e., an increasing tendency to use standard
was/were variants in contexts conforming to the
grammar of Standard English). The most plausible explanation for this
pattern is to regard it as a leveling process, a mechanism generally
involved in new-dialect formation (Trudgill,
1986), which involves the loss of minority and/or marked
variants and the adoption of fewer and more stable forms (Siegel, 1987). A closer look at gender-related
variation in 19th-century NZE reveals a cross-over effect, inasmuch as
it is the men who use more standard there were forms (and
appear to lead this change in progress, see Figure
1) and it is the women who use more plural agreement with
nonexistentials (and moreover, who are remarkably stable over the
entire period of NZE, see Figure 6). The
congruent directionality toward standardization in both environments
raises the question as to whether the two processes are related and how
we can explain the fact that they developed distinctly in 20th-century
NZE.
The relationship between singular concord in
existentials and nonexistentials
These data show that current NZE has high rates of singular concord
in existentials, whereas singular concord in nonexistentials is
virtually nonexistent. Concord operates differently with existentials
and nonexistentials, in that they are subject to distinct linguistic
constraints; for example, verb contraction has an enhancing effect on
singular concord in existential forms and an inhibiting effect on
nonexistentials. Such differences would suggest that the changes we
observe in this corpus result from underlyingly different mechanisms
(see, e.g., Kroch, 1989). On the other hand,
the two paradigms share properties, as well. For example, both
existentials and nonexistentials have a strong preference for singular
concord with past tense reference of be (at least in
noncontracted forms) and both display a (nonsignificant) tendency to
favor singular concord when noun and verb are distant. However, if all
is taken into consideration, the two paradigms are more distinct than
similar and are, with profit, studied as two separate phenomena and
subjected to different analyses, which is what we have done here.
Our analyses show that the two paradigms followed an unusual
trajectory during the formation and development stages of NZE.
Nineteenth-century NZE had both, yet by 1900, singular concord with
nonexistentials was marginal and remained practically nonexistent,
whereas singular concord with existentials began to increase again.
Thus, contemporary NZE has two quite separate tendencies, favoring one
and strongly disfavoring the other. This raises the question as to
whether concord in the two environments was connected in early New
Zealand English, which still had both, even though they were in the
process of decreasing. Was there a connection between the decline of
the two processes in the 19th century? Was usage of singular
be with plural subjects decreasing generally, regardless of
context, or did linguistic environment have an effect? To investigate
this, we calculated each Mobile Unit speaker's usage of singular
concord for both existentials and nonexistentials, and then examined
the degree to which these figures were correlated. The rationale is
that if there is a connection between decrease of singular
be in existentials and nonexistentials, then individuals with
low rates of one would also have low rates of the other. Figure 8 shows the result. Every point on the graph
represents a single speaker; filled points represent male speakers and
unfilled points represent females. Separate lines are drawn through the
male and female speakers; the solid line represents the line of best
fit through the females and the dotted line represents the line of best
fit through the males. Figure 8 shows a
highly significant correlation between usage of singular concord in
nonexistentials and existentials. The correlation is significant both
when calculated over the entire data set (Spearman's rho = .50,
p < .005) and when calculated separately for males
(Spearman's rho = .56, p < .01) and females
(Spearman's rho = .62, p < .05).
The relationship between existentials and nonexistentials (Mobile
Unit data only). Every point represents an individual. Filled points
represent males, unfilled points represent females. The solid line
represents the line of best fit through the females, and the dotted
line represents the line of best fit through the males (using a
non-parametric scatterplot smoother). This figure shows a highly
significant correlation between an individual's use of singular
concord in nonexistentials, and their use of singular concord in
existentials. (Males: Spearman's rho = .56, p < .01;
Females: Spearman's rho = .62, p < .05; Overall:
Spearman's rho = .50, p < .005).
We argue that this finding sheds new light on the change in progress
of existentials as outlined in Figure 1.
During the second half of the 19th century, singular concord of
be as a whole was being leveled out of New Zealand English.
The inception stage of NZE still had it, even though a (standard-like)
plural concord pattern became increasingly more frequent. Thus, we
argue, it is in this period that we observe a single change in
progress, namely, the reduction of singular concord with be,
even though existential environments, of course, had a favoring effect
on singular concord. Then, around the turn of the century, use of
singular be concord in nonexistentials all but disappeared
from NZE. Strikingly, although singulars were no longer used in
nonexistential constructions, they were still relatively frequent in
existential constructions (40–50%, Figure
1). At this point, the parallel decrease in both environments
reached the point where singular agreement with personal pronouns and
subject NPs de facto disappeared. Instead of disappearing as well,
there's/there was constructions began to
increase again, only to become very robust in contemporary NZE. How are
we to account for this phenomenon?
On the one hand, one might suggest sociopsychological and/or
stylistic explanations. For instance, World War I led to massive social
changes in New Zealand; particularly, New Zealand men became
increasingly mobile as a result of their involvement in the British-led
Allied Forces in Europe. This could suggest that they may have picked
up nonstandard there's/there was features
abroad and maintained this usage when returning home. However, although
Figure 1 documents a slight, yet
insignificant, trend for men to lead this change, it should be noted
that the varbrul analysis reflects an increase in singular
concord only for the women. As the increase in singular concord is more
significantly tractable for the women, it is unlikely that the return
of troops to New Zealand can account for the pattern.
A second consideration would be by nature stylistic. As we mentioned,
the year 1900 marks an overlap of two of our corpora used. Notably, the
two data sets were collected under different circumstances; the formal
nature of the MU data often involved an interviewer with an
unmistakable upper-class accent, and question–answer type
conversations that were not suitable for the elicitation of vernacular
forms. The IA corpus, on the other hand, was collected as part of oral
history projects; very often interviewer and interviewee were familiar,
and this obviously led to more informal and casual conversational
styles, and, by implication, to more vernacular variants. However,
whereas stylistic and social considerations may perhaps explain overall
differences in data from the two corpora, they offer no explanation as
to why we would find such a consistent nonlinear patterning. If
stylistic/social factors really account for the attested
vernacularization of NZE, then we would expect an increase in both
environments (existentials and nonexistentials) and are at odds to
explain the differential developments that occurred.
We are therefore tempted to offer a more language-internally driven
explanation, namely that the demise of singular concord with
nonexistentials more or less directly accounts for the increase in
singular concord with existentials. Because our statistical analysis
showed the two mechanisms to be related (Figure
8), we suggest that existentials developed separate agreement
properties when we was forms disappeared and the two paradigms
de facto ceased to be related. It is this reinterpretation of the link
between existentials and nonexistentials, we argue, that may be the
basis for explaining the reversal of the change in progress in the
early 1900s. There are thus two changes in NZE singular concord: (1)
conjoined standardization in the 19th century, which ended with the
loss of singular concord with personal pronouns and plural NPs at
around 1900; and (2) the subsequent increase in singular concord with
there constructions in the 20th century.
A cross-dialect perspective of singular agreement
The last point we would like to address concerns the wider
implications of our study. How does subject–verb concord in NZE
compare with the patterns reported for other varieties in England, the
United States and Canada, or the Falkland Islands? Are there parallels
between these dialects? Or, does agreement align differently in
each?
We suggest that there are parallels and differences between dialectal
patterns of subject–verb concord. While some principles seem to
hold generally, other features are subject to variation and
differentiation. Our study uncovers several findings that are attested
in every dialect that has been subject to analysis: (1) singular
agreement with nonexistentials is much more likely to operate when
verbs have past reference (the dogs was > the dogs
is); (2) existentials are much more likely to exhibit singular
agreement than are personal pronouns and plural NPs; and (3) linguistic
factors such as tense, contractedness, and modifier type have an effect
on the frequency of singular concord. In these respects, NZE is
perfectly in line with practically all other forms of English. However,
we also see a number of minor differences that set New Zealand English
apart from other varieties. One of these concerns the effect of
modifier type. Table 2 provides a comparison
of different hierarchies in Canadian English (Meechan & Foley, 1994), York English (Tagliamonte, 1998), New Zealand English (Britain
& Sudbury's 2002 results, labeled NZ
English 1), Falkland Islands English (FI English, Britain & Sudbury, 2002), and our results
(labeled NZ English 2).
Even though subject–verb concord is subject to implicational
gradations that hold for all varieties (e.g., a ‘no’
quantifier is more likely to have singular forms than bare NPs or
adjectives), the individual ordering of these constraints differs. This
may be a function of the idiosyncrasies and sizes of the different data
sets analyzed, or it may suggest that each dialect has somewhat
idiosyncratic constraint hierarchies.
