This section adds the 1930s figures from B-Brown and B-LOB to those published in Leech et al. (2009: 283) in order to complete the Brown-based survey of the major trends in the use of the nine core modals and the two marginal modals need and ought (to) in the 20th century. Since the publication of Leech et al.’s (2009) study, Leech
& Smith have published follow-up work on modals and selected semi-modals in B-LOB, using a part-of-speech tagged10 version of the corpus and giving normalised
10. The tagset used was CLAWS, C8: see Leech and Smith (2005: 92).
Free ebooks ==> www.ebook777.comCross-variety diachronic drifts and ephemeral regional contrasts 129 frequencies per million words (Leech & Smith 2009: 187–192; Leech 2011; Leech 2013). As a comparable part-of-speech tagged version of B-Brown is not available yet, the present comparison between B-LOB and B-Brown is based on the untagged data and on absolute rather than normalised frequencies. This procedure has the advantage of (a) independently verifying the findings published by Leech & Smith, and (b) allow- ing direct comparison of B-Brown and B-LOB.
For the British branch of the extended Brown family, the present study fully con- firms Leech & Smith (2009) and Leech (2011, 2013). Numerical differences between the B-LOB figures in Leech & Smith’s work and those given in Table 5 below are almost negligible, which testifies to the high accuracy of the tagger in identifying modal verbs.
Table 5. Nine core modals and two marginal modals in B-LOB, LOB and F-LOB11 B-LOB
(1930s) LOB
(1961) Change %
B-LOB→LOB F-LOB
(1991) Change %
LOB→F-LOB Change % overall
would 2,673 3,032 ****+13.4% 2,682 ****–11.5% +0.3%
will 3,055 2,822 *–7.6% 2,708 –4.0% ***–11.4%
can 2,039 2,147 *+5.3% 2,213 +3.1% **+8.5%
could 1,433 1,741 ****+21.5% 1,767 +1.5% ****+23.3%
may 1,702 1,333 ****–21.7% 1,100 ****–17.5% ****–35.4%
should 1,486 1,301 **–12.4% 1,148 **–11.8% ****–22.7%
must 1,265 1,147 –9.3% 814 ****–29.0% ****–35.7%
might 713 779 *+9.3% 640 ***–17.8% –10.2%
shall 475 355 ***–25.3% 200 ****–43.7% ****–57.9%
ought (to) 135 103 –23.7% 58 ***–43.7% ****–57.0%
need(n’t) 94 76 –19.1% 44 **–42.1% ****–53.2%
Total 15,070 14,836 –1.6% 13,374 ****–9.9% ****–11.2%
(LOB and F-LOB figures from Leech et al. 2009: 74, 283; significances [log likelihood]: * = p < 0.05,
** = p < 0.01, *** p < = 0.001, **** p < 0.0001)
There is straightforward diachronic continuity from the 1930s to the 1990s in several cases. The rare modals shall, ought and (auxiliary) need show continuous decline from B-LOB to F-LOB, as do medium-frequency must, may and should, and high-frequency will. Discontinuous developments, on the other hand, are documented for would and might. Here the (statistically highly significant) decline in the 1961 to 1991 time window recorded in Leech et al. (2009: 283) reverses an equally significant previous trend. In the absence of a plausible motivation for these ups and downs, we
11. The corpus searches in B-LOB were carried out by Mirka Honkanen, whose help is gratefully acknowledged.
www.ebook777.com
Free ebooks ==> www.ebook777.com
130 Christian Mair
must assume topic- or genre-related fluctuation rather than a directed diachronic trend.
One fact which deserves note is that the rather drastic overall decline in frequency of the core modals as a category (–9.9 per cent) which Leech et al. noted for the thirty- year period from 1961 to 1991 was preceded by a much more modest – and statistically insignificant – decline (–1.6 per cent) in the preceding thirty-year window.
For BrE, it is possible to compare developments in the spoken and written lan- guage during the second half of the period covered by the extended Brown fam- ily. In their analysis of the DCPSE (the Diachronic Corpus of Present-Day Spoken English), Bowie et al. (2013: 80, 93) show that must, may and shall experience signifi- cant declines in informal face-to-face conversations, whereas can and might record increases. There are no statistically significant findings for could, should, will and would in this register.12
However, as Bowie et al. also show, the fact that the frequency of a particular modal remains stable in the baseline register of spontaneous conversation does not prevent significant diachronic shifts in other registers. The modal should, for instance, shows significant decline in formal face-to-face interaction and prepared speech, but an increase in the DCPSE category “parliamentary language” (2013: 80). Not unex- pectedly, written BrE shows a steady decline for should from B-LOB to F-LOB, which thus follows the trend in formal speech.
There are three modals – must, shall and may – for which Bowie et al. are able to show a decline independent of genre factors:
The modals must, shall and may are the only ones which decline as a proportion of the core modal set when all text categories are considered together. They show substantial declines of 51% (± 11%) for must, 45% (±19%) for shall, and 36%
(±19%) for may. In [Aarts et al. 2013], we performed a within-set analysis of the same core modals in the British written English corpora LOB/F-LOB and compared these with DCPSE. In LOB/F-LOB the comparable figures are declines of 22% (±8%) for must, 38% (±13%) for shall, and 9% (±7.5%) for may. Moreover, the results for must and may were found to be statistically separable. That is the relative declines for those two modals are significantly greater in the spoken corpus, suggesting that spoken English is “leading the way” for these changes.
(2013: 87–88)
For two of these three modals, namely may and shall, a steady decline is already recorded from B-LOB to LOB; for must, on the other hand, the statistically significant decline does not set in until the 1960s.
Table 6 presents the findings from the American members of the Brown family.
Of particular interest are the previously unpublished figures from B-Brown, because
12. The marginal modals need and ought (to) are not considered.
Free ebooks ==> www.ebook777.comCross-variety diachronic drifts and ephemeral regional contrasts 131 they help us distinguish the major developmental trends affecting both varieties from the more short-term fluctuations encountered in one or the other. For AmE, we have no spoken corpus comparable in design to the DCPSE, but it will be useful to compare findings from the small corpora of the Brown family to those obtained from COHA, a very large historical corpus of AmE.
Table 6. Nine core modals and two marginal modals in B-Brown, Brown and Frown13 B-Brown
(1930s) Brown
(1961) Change %
B-Brown→Brown Frown
(1992) Change %
Brown→Frown Change % overall
would 2,412 3,053 ****+26.6% 2,868 *–6.1% ****+18.9%
will 2,606 2,702 +3.7% 2,402 ****–11.1% **–7.8%
can 1,718 2,193 ****+27.7% 2,160 –1.5% ****+25.7%
could 1,332 1,776 ****+33.3% 1,655 *–6.8% ****+24.3%
may 1,357 1,298 –4.3% 878 ****–32.4% ****–35.3%
should 1,037 910 **–12.3% 787 **–13.5% ****–24.1%
must 955 1,018 +6.6% 668 ****–30.4% ****–29.9%
might 626 665 +6.2% 635 –4.5% +1.4%
shall 289 267 –7.6% 150 ****–43.8% ****–48.1%
ought (to) 111 69 **–37.8% 49 –29.0% ****–55.9%
need(n’t) 49 40 –18.4% 35 –12.5% –28.6%
Total 12,492 13,991 ****+12.0% 12,287 ****–12.2% –1.6%
(Brown and Frown figures from Leech et al. 2009: 74, 283; significances [log likelihood]:
* = p < 0.05, * *= p < 0.01, *** p < = 0.001, **** p < 0.0001)
There are some obvious parallels between BrE and AmE. As in the British cor- pora, the rare modals shall, ought and (auxiliary) need show continuous decline from B-Brown to Frown, and so do medium-frequency must, may and should. There are minor differences in the rates and levels of decline. For example, shall is comparatively infrequent in the American corpora even in the 1930s, whereas must experiences an unexpected temporary increase between B-Brown and Brown. For the high-frequency forms will, would, can and could, there is fluctuation rather than long-term directed change (as could mistakenly have been inferred from looking at the developments between either the pair B-Brown-Brown or Brown-Frown alone). The figures for would, in particular, are “all over the place”: a rise followed by a fall to produce the
13. The corpus research on B-Brown was carried out by Rachel Hill, whose help is gratefully acknowledged.
www.ebook777.com
Free ebooks ==> www.ebook777.com
132 Christian Mair
appearance of stability over sixty years in the British data; a rise only partially offset by a subsequent fall in AmE. This squares rather well with a similarly heterogeneous distribution found for this modal in the DCPSE, where Bowie et al. (2013: 80) note stability in eight out of ten spoken genres, but a robust rise in broadcast discussions and an equally strong dip in prepared speech.
The generalisation that modals are on the decline as a category, which was for- mulated on the basis of data from the original Brown quartet in Leech et al. (2009),14 cannot be upheld for B-Brown, Brown and Frown. A 12 per cent rise from the 1930s to the 1960s is offset by a 12.2 per cent decline in the following period. The two succes- sive and highly significant 30-year shifts in frequency do not add up to a directed and statistically significant 60-year diachronic trend.
However, a decline of modals as a category can be attested from other corpora.
As B-Brown was not available at the time of his writing, Leech (2011: 553) analysed modals in COHA, the Corpus of Historical American English, to document trends in AmE in the 20th century. With regard to composition, COHA is a mixed-genre corpus sourcing material from newspapers, magazines/periodicals, other non-fiction and fic- tion and thus broadly similar in design to a Brown family corpus. Leech reports the decade-by-decade shifts for the total frequency of modal verbs in the 20th century as presented in Table 7.
Table 7. Frequency of modal verbs in COHA (normalised: pmw, Leech 2011: 553) 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 1910–
2000s Totals
(pmw) 15,273 14,295 14,121 14,168 14,142 13,903 13,789 13,025 12,908 12,076
% change from prev.
decade
–6.40
*** –1.22 +0.33 –0.18 –1.69 –0.82 –5.61
*** –0.90 –6.45
*** [–20.9]
***
NB: The figure for the overall decline 1910–2000s is not given by Leech but was calculated using his figures.
As can be seen, Leech records a statistically significant decline from the 1910s to the 1920s, followed by five decades of stability, and then decline again from the 1970s.
14. This is Leech et al.’s comment on the compound frequencies in the British and American corpora: “[T]he general trend is a significant loss of frequency: the decline is 10.6% for the class as a whole, while the loss for individual modals varies between 2.2% (could) and 43.5%
(shall)” (2009: 73).
Free ebooks ==> www.ebook777.comCross-variety diachronic drifts and ephemeral regional contrasts 133 The overall downward trend for the 20th century (rightmost column) is significant, as is the decline from the 1930s to 1990s, the time window covered by the Brown family (p < 0.0001).
Which evidence should we trust more – three small but carefully matched cor- pora, which show ups and downs in two consecutive thirty-year intervals levelling out to stability, or a less rigidly structured mega-corpus, which records a statistically significant decline of the modals for the same time period and the 20th century as a whole? There is the additional twist that the decline in the overall frequency of the modals documented in COHA is largely due to a dramatic dip in one single decade (1970s to 1980s) – a decade which happens to be included in the thirty-year period (Brown to Frown) for which a decline is also recorded in the small corpora. The rela- tive merits of small and large corpora in this type of diachronic study was the topic of a recent debate between Neil Millar and Geoffrey Leech. Millar (2009) argued that the Brown family corpora were small by contemporary standards and that diachronic research based on them also suffered from the inflexible sampling points spaced out over 30 years. For his study of modals in 20th-century AmE, he therefore used the TIME Magazine Corpus (http://corpus.byu.edu/), which contains all issues of Time magazine from 1923 to the present. At 100-million-plus words it is much larger than the Brown family and allows year-by-year documentation of diachronic trends. As some of his findings, particularly on the modal may, were different from those pub- lished in Leech and Smith (2009), he argued that their analysis was flawed because of the lack of sufficient data. In his response, Leech (2011) conceded the problems arising from the inflexible sampling points but showed that small corpora covering several genres provide more representative results than large corpora such as the one used by Millar, which was drawn from a single text-type or, in Millar’s case, even a single publication.15
In fact, internal analysis of Brown-type corpora reveals drastic genre-dependent variation in the frequency of modals and semi-modals, and this is certainly an argument in favour of the continued use of small corpora which are both stratified by genre internally and carefully matched for genre for comparison across corpora – whether we are dealing with the history of BrE and AmE or the New Englishes.
To get a hold on the slippery genre determinants of modal usage, it is useful to separate the different genres within the corpora, as is done for B-Brown and B-LOB in Table 8.
15. For further discussion of the relation between regional variation and genre in newspaper language, see Noởl & Van der Auwera, this volume.
www.ebook777.com
Free ebooks ==> www.ebook777.com
134 Christian Mair
Table 8. Frequencies of modals according to genre in B-Brown and B-LOB (per million words)16
B-Brown B-LOB
Sub-corpus (number of
2,000-word samples) freq. (pmw) ranking freq. (pmw) ranking
Press (88) 12,195 3 13,270 3
Gen. prose (206) 11,858 4 15,019 2
Learned (80) 12,646 2 12,749 4
Fiction (126) 13,092 1 16,267 1
Corpus average (500) 12,355 – 14,65816 –
It is instructive to compare Table 8 with the corresponding one in Leech et al.
(2009: 75), who compare LOB and F-LOB (i.e. BrE in short-term diachrony), because this again highlights the extent of variability in modal usage across genres. For LOB, they obtain the ranking Fiction > Press > General Prose > Learned, and for F-LOB Fiction > Learned > Press > General Prose. Interestingly enough, this latter ranking – obtained for the most recent British material – happens to be the one recorded here for the oldest American corpus (although the frequencies, of course, are very different).
The highest genre-specific frequency in B-Brown is 6.0 per cent above the corpus aver- age, and the lowest 4.0 per cent below it, whereas the spread is much wider in B-LOB, with the highest exceeding the average by 11.0 per cent and the lowest undershooting by 13.0 per cent. Neither the different genre-rankings nor the variations in the spread from the average allow of an easy explanation. The only two constants, both in the present study and in the corresponding table in Leech et al., are that modals are most frequent in fiction (rank 1 in B-Brown, B-LOB, LOB and F-LOB) and that differences between BrE and AmE are smallest in the Learned genre, reflecting the high degree of international standardisation in scholarly writing and publishing. Therefore, Collins’
(2014) focus on fictional texts for his pioneering study of the history of modals in AusE was a good choice – although, of course, other genres will have to be investigated in future to complete the picture.
As the preceding discussion shows, historical corpus studies of BrE and AmE can fall back on an abundance of material. This is not yet so for most New Englishes, and historical research on these varieties is just beginning. However, in the case of Indian
16. The corresponding figure in Leech (2011: 551) is 14,655. The minimal difference is evidence of the accuracy of the part-of-speech tagging.
Free ebooks ==> www.ebook777.comCross-variety diachronic drifts and ephemeral regional contrasts 135 English we have corpus material at least for a diachronic micro-study, namely the 1978 Kolhapur corpus and ICE India, with texts sampled in the 1990s. Table 9 presents the findings for must and shall from the Kolhapur corpus and the written texts of ICE-India:17
Table 9. Must and shall in Kolhapur and ICE-India Kolhapur ICE-India (written, normalised
frequencies pmw)
must 766 780
shall 364 318
(significance: p < 0.05, not significant)
On this evidence, a robust trend, attested in other varieties, for must to decline in frequency has bypassed Indian English in the recent past. More diachronic time-depth would be needed in order to confirm this exceptional development.