Sesquiotica

This is the full version of my paper on “an historic” and “an historical.” Apologies for the poor fit of some of the tables. They look nicer in the PDF version.

Introduction

One of the most regular and inflexible rules of English is the one governing which version of the indefinite article to use in a given context. It is a useful thing to have an understanding of the rule, and it would take less than an hour to learn a habit of choosing according to the sound of the following word: a before a consonant, as in habit, but also before a consonant sound written as a vowel, as in useful; an before a vowel, as in understanding, but also before a silent consonant (inevitably h) followed by a vowel, as in hour. Although in some dialects a is used before vowels as well, this usage is considered nonstandard and is generally looked down upon (notwithstanding which it has occasionally been predicted that this will be the ultimate use everywhere – see, for example, the editor’s note following Bolinger 1975). An before a consonant would be considered a mark of a nonnative speaker.

There is, however, a salient exception. Before a few words that begin with [h], and most notably historic and historical, an may often be seen and heard used in place of a, even by people whose dialect does not delete the [h]. It seems to have gained an air of greater correctness and formality in many circles. Although a is more common, an is widely seen, especially before historic. Google searches, worldwide and for .ca (Canadian) domain names only, give an indication:

	a historic	an historic	ratio a:an	a historical	an historical	ratio a:an
Google global	2,790,000	1,310,000	2.13	24,200,000	1,280,000	18.91
Google site:.ca	178,000	128,000	1.39	531,000	360,000	1.48

The Canadian government’s websites (all sites in the .gc.ca domain) prefer a to an for historic by only 1.28:1, with 4,570 and 3,570, respectively, and for historical by 1.83:1, with 8,280 and 4,530.

Nor is this a casual matter of personal choice; it is much debated, and positions are often firmly held. There is no shortage of people who will assert quite flatly that “an historic is actually the correct pronounciation” (Urban Dictionary 2004) and even counsel those who prefer a to “look it up” (Yahoo Answers U.K. & Ireland 2007) – ironically, given that current British and American usage manuals almost without exception either explicitly prefer a or at least allow it. Some speakers will aver that “it sounds better to say ‘an historic’” (Yahoo Answers 2006; see also Opinion L.A. 2007); some will simply say “there’s a case to be made that an is the suitable article before historic” (Opinion L.A. 2007). Many will use it because they are certain that it is correct or more formal; others will chose it because, being uncertain, they choose the more marked version on the assumption that it would not be used by others if it were not correct. As Bolinger (1975) quotes Ralph Long as saying (in a personal communication), there is a tendency among “people who really know little or no English grammar…, when in doubt between two constructions, [to] pick the less usual and presumably more elegant.”

Those on the other side of the issue declare an historic to be “pretty much a sherry-sipping, bowtie-wearing thing” (City Comforts 2006) or “pedantic, fussy, and patronising” (Yahoo Answers 2006), fume “I hate that ‘an’ preceding ‘historic’ … it seems awfully pretentious” (Opinion L.A. 2007), or simply flatly declare it wrong: “Do you live in an house? I didn’t think so. A historic” (Walsh 2006). Some usage guides allow either usage, but the trend among authorities appears to be in favour of a. As Fee and McAlpine (1997) put it, “British usage guides are recommending against the unnecessary an. It is probably time for Canadians to let it go too.” And yet many seem loath to do so.

There are four questions that deserve answers in this regard: First, how did this state of affairs come to be? Second, what in fact do most people consider more correct and more formal? Third, why is this the case? And fourth, what is the trend for the future for this usage?

Background

The dispute over which article to use with historic and historical is not new, although the restriction of the dispute largely to those two words is of more recent date. Hillhouse (1928) quotes a piece titled “Humble Petition of the Letter H” from the Grub-street Journal of January 24, 1733–4. In it, the letter H “begs leave to remonstrate against the prevailing custom of authors or printers, or both, who always set the particle An before a word that begins with H: by which method they injuriously deny that he is any letter at all, since, to be sure, they will not call him a vowel.” H continues by asserting that it is already “by a good custom settled for speaking” that words in which H is pronounced are preceded by a; “If men will write An house, an horse, an high-lander, they ought to read so, too. But if it be ridiculous to read so, it must be as ridiculous to write in this manner.” The case, then, was that although pronunciation had long since shifted to restore or add the pronounced [h] which had been dropped under French influence, printers and writers still often preferred the traditional usage (as they did with many points of spelling). Not always, however; Hillhouse observes that, although an before pronounced h could be found with many words in instances from some writers well into the 18^th century, a had long since become the established norm, and had been appearing in print since the 16^th century. Mark Liberman (2004) tracked usages of an hero using the literary database lion.chadwyck.com and tracked the death dates of the more than 60 authors who used it; he found that the first three authors cited were in the last half of the 17^th century, that numbers increased to a peak around 1800, and that they then dropped sharply to 1900. Thus there seems to have been a vogue, and one that came about not in concert with the French influence but rather more in line, perhaps, with the late-18^th-century flush of prescriptivism (however, the uprising arc before 1800 may also reflect the composition of the database).

The use of an came to be restricted to h-words with an unaccented first syllable, for example historian and historic. But even that had come to deprecation, though not disuse, by the late 19^th century. Hillhouse quotes the 1888 New English Dictionary: “this is all but obsolete in speech, and writing a becomes increasingly common in this position.” He adds an admonition from the noted prescriptivist H.W. Fowler in his 1926 Dictionary of English Usage: “now that the h in such words is pronounced the distinction has become pedantic, and a historical should be said and written.”

The door was not closed on the issue, however. In 1929, Louis N. Feipel published a survey of 300 books, divided equally between American and British authors, examing their use of the indefinite article before h and vowels that are preceded by glides such, as “long u” [ju]. He found an assortment of instances of an before h in monosyllables and words accented on the first syllable – 11 each from American and British books. He found rather more instances when he turned to words starting with [h] not accented on the first syllable. The word most commonly preceded by an was, in fact, hotel – Feipel notes that “‘an hotel’ preponderated markedly over ‘a hotel’; but strangely enough, of the many ‘an’ instances only one was by an American writer.” Next after hotel was historic(al) – Feipel treated the two words as one. Here Feipel found usage “evenly divided between ‘a’ and ‘an,’ as also between British and American writers.” (The actual instances listed numbered as follows: a: 7 British, 4 American; an: 4 British, 4 American.) Following historic(al) was heroic, for which an preponderated, but especially among the British. Other words for which an was more common than a included hallucination, hysterical, horizon, hypothesis, habitué, hereditary, hermaphrodite (itic), hermetical(ly), and several that had only one instance each. On the other hand, a preponderated for hypnotic, harmonious (harmonium), Havana, and several words with one instance each, and there were also a few words that were evenly split.

Feipel’s article drew some responses. One (Byington 1929) noted that much of the variation in style could be attributed to the proofreaders and copy-preparers at the various publishing houses, and that they are more likely to be dogmatic and perhaps tradition-bound than the average user; the next (Palmer 1929) declared “it has long seemed natural to me to use an before an unaccented h. A historical, and a hypothesis offend my ear.” These were followed by a note from the editor (Kenyon 1929), who noted briefly the history and conventions and declared that the inconsistency was not surprising.

The inconsistency persisted for the following half century – but specifically with historic(al). In 1975 Dwight Bolinger declared an historical to be “another presitigious contagion” that was “spreading fast in both print and sound, these days.” In response to Bolinger, Bollard (1979) surveyed material collected by the pronunciation editors at the G. & C. Merriam Company. He found that an preponderated in the speech sampled, especially for historian, historic, and historical, by margins of 22:1, 28:3, and 24:1, respectively, and that it was preferred by smaller margins with several other h words. The totals in writing samples bore out the same result, with larger numbers of instances but smaller ratios (63:25, 85:51, and 194:98, respectively), and also with other words such as habitual, hereditary, and hallucination. More telling was the breakdown of the pronunciation variants: for historian, historic, and historical, in total, of 65 instances recorded, 25 pronounced the [h]. This means that the tradition prevailed even in the face of phonological contradiction of its original justification. There is also the matter of how many of the [h]-less instances were said by people who would say the [h] in the absence of the indefinite article. This practice has evolved as a “rule” that some users hew to. Bolinger adverts to this when he categorizes “h-droppers” in three groups: those who always pronounce the h, even with an (“the true phony h-dropper”); those who never pronounce the h (“the sincere h-dropper”); and those who drop the h just after an: “He writes an historical and says an ’istorical, but elsewhere does not spare his aspiration in the historical record, no historical justification, by historical methods. He is half-phony because he stands a rule of English on its head, which is that what follows determines the shape of the article; the article does not determine the shape of what follows.”

In the 21^st century, an historic is still seen – and widely thought correct – and, even more notably, a historic is thought by many to be wrong. The situation is such that the more descriptivist New Fowler’s Modern English Usage (Burchfield 1996, 2) allows the choice of a or an as a matter of personal preference. Most modern style guides and expert writers on the subject disagree. Bill Walsh, who maintains a site for copyeditors called The Slot and is the author of a few books on English usage, surveyed (2004) several style guides and found that, while the London Times called for an in its stylebook (for hotel and heroic as well as historic), and two American dictionaries and two American usage guides allowed the user’s choice, the remainder of American guides sided firmly with a: Garner’s Modern American Usage, Patricia T. O’Conner’s Woe Is I, The Chicago Manual of Style, The Associated Press Stylebook, The United Press International stylebook, the Washington Post stylebook, the New York Times stylebook, the USA Today stylebook, and the U.S. News & World Report stylebook. One person taking the other side in debate with Walsh cited The Correct Word: How to Use It by Josephine Turck Baker: “when h is aspirated, a is required, unless the accent is on the second syllable, when an is used; as ‘a history;’ ‘an historian.’” However, Baker was writing in the first decades of the 20^th century – at a time, in fact, when most authorities had already begun counselling users to prefer a in such contexts. Canadian style guides likewise counsel a rather than an (see Editors’ Association of Canada 2000, 211; Tasko 2005, 91; Fee and McAlpine 1997, 1).

Has an use peaked? Has it described, in the broad view of history, an arc like the one that Liberman discerned for an hero? A check of the same database as Liberman used, lion.chadwyck.com, shows parallel quantities of usage for both versions persisting from the 18^th century to the 20^th, with an usage holding about a 10:7 ratio over a usage. Authors with no birth or death dates listed (which in this database are usually living authors with recent works) are skewed to a by a 10:4 ratio, which may indicate a change in progress. However, the total number of authors cited, 49 for an and 34 for a, is too small to be conclusive.

A search of some Canadian, American, and British news media websites finds the following results:

	a historic	an historic	ratio a:an
Toronto Star	159	42	3.79
Globe and Mail	766	142	5.39
National Post	353	39	9.05
Macleans	40	15	2.67
CBC	1,820	492	3.70
CTV	1,390	298	4.66
New York Times archive 1981–	7,529	841	8.95
New York Times archive 1851–1980	8,138	1,575	5.17
Wall Street Journal	836	236	3.54
London Times	1,230	1,610	0.76
London Telegraph	1,820	1,040	1.75

	a historical	an historical	ratio a:an
Toronto Star	77	16	4.81
Globe and Mail	258	38	6.79
National Post	90	17	5.29
Macleans	23	0	n/a
CBC	562	112	5.02
CTV	362	29	12.48
New York Times archive 1981–	5,291	305	17.35
New York Times archive 1851–1980	11,381	1,687	6.75
Wall Street Journal	429	82	5.23
London Times	533	249	2.14
London Telegraph	1,170	537	2.18

Only one site has an more than a for historic, the Times of London, which calls for it in its style guide (and even still it has a nearly three-quarters as often as an). Its London competitor the Telegraph has a nearly twice as much as an. For historical, preference for a is universal though not absolute. Every North American news outlet surveyed preferred a for both words by a notable margin. And it is worth remembering that many of the instances will have been in quotations (though the search results show that some of the usages are by the organizations’ own writers).

A search of the same outlets for rations of a to an for habitual, hysterical, hotel, and hero finds interestingly varied results (n/a means that there were no instances of an at all):

	hysterical ratio	habitual ratio	hotel ratio	hero ratio
Google global	4.44	2.94	37.98	36.44
Google site:.ca	0.73	1.25	130.71	343.80
Goole site:.gc.ca	1.55	0.42	143.83	462.86
Toronto Star	n/a	n/a	202.50	n/a
Globe and Mail	n/a	n/a	n/a	n/a
National Post	n/a	n/a	n/a	n/a
Macleans	n/a	n/a	n/a	n/a
CBC	3.78	24.33	48,600.00	n/a
CTV	n/a	n/a	1,052.00	n/a
New York Times archive 1981–	9.07	18.77	2,641.33	7,116.00
New York Times archive 1851–1980	11.38	2.23	135.40	113.61
Wall Street Journal	n/a	24.00	536.67	n/a
London Times	4.08	1.42	49.30	n/a
London Telegraph	5.25	3.03	284.04	n/a

We can see that hysterical and habitual still get a fair amount of an usage in some quarters, but none at all in most Canadian news outlets surveyed, while hotel gets very little an usage and hero quite nearly none – but not absolutely none. Interestingly, the London Times also has a heavy preponderance of a hotel in spite of its style guide’s prescription. Most striking, perhaps, is the prevalence of an with habitual on Government of Canada websites – due to its standard use in legislation – and the prevalence of an with hysterical on .ca sites, something that might reward further study in a future research effort. Given its absence in the usage of Canadian media outlets, this latter would seem to be an anomaly. (We should also remember that .ca domains are used only by a subset of all Canadian websites.)

While a is winning, however, an still has a strong presence with historic and historical, and to a generally lesser degree with a few other similar words. Style guides tend to focus on historic and historical in this issue; the general consensus is that these words are the strongest survivors: “Nowadays the use of an before h survives primarily before the words historical and historic” (ITP Nelson Canadian Dictionary of the English Language 1998, 1). Certainly it prevails in terms of absolute numbers; while the a:an ratios may be similar for some other words in some sources, their frequency of usage is much less – typically two to three orders of magnitude less.

Why has an persisted with historic and historical? No doubt there has been some effect of linguistic ideology (see Wollard and Schieffelin 1994 and Kroch and Small 1978 for general discussions of the topic) – as Silverstein (1979) defines it, “sets of beliefs about language articulated by users as a rationalization or justification of perceived language structure or use”; Irvine (1989) calls it “the cultural system of ideas about social and linguistic relationships, together with their loading of moral and political interests.” The ideological aspect is manifested especially in the tone of some of the debate: the an usage is strongly associated with a British-style prestige model, one viewed by some as the truly correct model (deviation from which offends the ear) and by others as intolerably elitist (“sherry-sipping,” “patronising,” and “pretentious,” to reprise three quotations from above). Ideology tends to override other factors and can be used to justify many an exception – and many a vociferous exception to that exception! As Milroy (2004) says, language ideologies are typically “historically deep-rooted and thoroughly naturalized – hence their resistance to analysis or argument.” A usage may be justified with reasoning that may not reflect the speaker’s phonological reality – as witness this statement from Yahoo Answers 2006: “the h in ‘historic’ is not really acting like a consonant. It forms a sort of dipthong with the I.”

Another possible source of the current state of affairs is the context in which the word historic is often seen. It happens to be a word that is often associated with events that are, well, historic, and thus formal and exceptional. In 1949, Ralph H. Lane noted that “the American likes historic when the word denotes prestige or age, and he applies it somewhat indiscriminately, because of a national affinity for the adjective which dignifies events and objects connected with his rude forefathers.” After surveying the Washington Post for the first half of 1948, he observed that “at the present time historic (especially when it appears in the press) predominantly denoted prestige.” When we consider the importance and exceptionality that can be associated with historic, we may imagine that a more conservative “harking back to olden days” may be in operation some of the time.

Problem and hypothesis

The information we have just reviewed gives good historical, accessory, and anecdotal information about the matter at hand, but no detailed survey of actual current use. I therefore set out to determine what percentage of speakers in Canada today consider each of the usages correct or incorrect: a versus an with historic and historical. I also wished to determine whether there is a relationship between perceived formality and perceived correctness in these usages; I wished to test for an effect of a possible overall set of linguistic ideological beliefs; I wished to test the extent to which usage is determined by pronunciation or non-pronunciation of [h] at the beginning of words; and I wished to find out whether preference for one or the other related in any significant way to demographic details such as age or educational background.

I hypothesized that perceived formality would be a factor in choice of an over a but also that an would be found more formal even among those who thought it incorrect. I also hypothesized that there would be a relationship between preference for an and preference for certain favoured prescriptivist rules. As well, I hypothesized that there would be many people who pronounce the [h] who nonetheless use an with historic. And I hypothesized that respondents’ ages would have significant relationship with their views on the correctness of a and an with historic and historical.

Methodology

I developed a web-based survey that asked for respondents to rate 20 sentences on formality and correctness and to give demographic information. See Appendix 1 for details of the survey used. The stimulus sentences and demographic questions were presented all on one page. This meant that respondents could change their response to a question at any time while filling out the form, up to when they clicked on “Send.” Respondents were randomized to one of two forms by means of the last digit of their postal code. The stimulus sentences on the forms were matched, in most cases offering two variants on a specific grammatical feature, and in some cases offering the same sentence on both forms.

Participants were solicited by means of email. The emails were distributed to four groups: a) employees of MediResource Inc., a Toronto-based web health information company; b) members of the Toronto Mendelssohn Choir; c) members of the Church of the Holy Trinity, which is in Toronto; d) acquaintances of my father, who lives in Cochrane, Alberta, but writes a weekly column that is read by email and on the web by people across Canada and elsewhere. The last group was the largest and also the oldest on average; its presence resulted in the median age being higher than it would otherwise have been. (The average age could have been reduced, and the response pool increased, by soliciting responses from two other groups with which I have connections, but these groups consisted of linguistics students and professional editors, and I felt that either of these groups would skew the results due to their unusual awareness of and focus on matters of English usage.) The emails also asked respondents to forward the email to family and friends to get their participation as well. In total, 214 responses were received: 103 for form 1 and 110 for form 2, and one discarded as an evident accidental duplicate of the immediately preceding one, probably by double-clicking rather than single-clicking the “Send” button.

Personal experience has indicated to me that when one asks people directly about a point of usage, they do not always give answers that reflect their actual usage; sometimes they are unable to remember, and sometimes they say what they think the questioner wants to hear. Thus, rather than asking directly whether a or an is correct before historic and historical, and how formal the usage is, I presented a set of sentences that they could rate on correctness and formality, without focusing on the grammatical feature of interest to me in the sentence, and I spaced the questions of most interest suitably far apart in the form so as not emphasize the focus. I used two forms so that I could test variants without calling attention to the variation being tested. Form 1 had a sentence containing a historic (“This is a historic occasion”) and a sentence containing an historical (“They conducted an historical survey”); form 2 had the same sentences but with the a and an reversed. Since I didn’t want to ask directly how the respondents pronounced words starting with h, because I didn’t want to turn their attention directly to what I was trying to find out, I used a sentence with an hotel to test this (“There was an hotel on the other side of the river”), reasoning that those who did not pronounce the h would generally find this correct and those who did would generally find this incorrect. The remaining 17 sentences were chosen to manifest “correct” and “incorrect,” and more and less formal, forms of certain usages. Some of the sentences used hinged on points of usage that have a certain prescriptivist shibboleth value: split infinitive, sentence-ending preposition, hopefully as a sentence adverb, and a few others. Some items were included with the expectation that the data gathered on them may be useful for future investigation of specific points of usage.

Some items were identical on both forms so as to provide stable points of comparison or to give a comparatively formal, informal, correct, or incorrect item for the sake of comparison (e.g., for informal, “You want me to do what?”). For most of the items, there were two variants, one expected to be thought more formal and/or correct, the other less formal and/or incorrect. The distribution of these items was balanced between the forms with the intent of making each form seem roughly equal. In the final results, the mean formality value for all items on form 1 was 2.6375, the total “no” (incorrect) was 911, and the total “yes” (correct) was 1036. For form 2, the mean formality was 2.664, the total “no” 861, and the total “yes” 1234. The difference in mean formality was not statistically significant; however, a chi-square test found that the difference in total correctness values, about 57 from predicted values (i.e., the values that would obtain if the overall proportions between rows and columns held for each individual item), was significant:

	form 1	form 2
incorrect	911	861
correct	1036	1234
predicted values
	853.56	918.44
	1093.44	1176.56
	p=0.0003

It is not known what effect, if any, this overall difference had on the judgements of individual items. The historic/al items naturally affected this total, but even with them excluded the difference was statistically significant, differing by about 41 from expected values:

	form 1	form 2
incorrect	813	786
correct	944	1096
predicted values
	772.04	826.96
	984.96	1055.04
	p=0.006

The data from all responses were aggregated in tabular form and subjected to a variety of analyses to find relationships between sets of responses. Because of the small range of possible choices (five values for formality and effectively two values for correctness, since the number of “uncertain” choosers was too small to be useful statistically, so they were excluded), correlation and ANOVA tests were not considered suitable; chi-square tests were preferred for the correctness questions, while Student’s t-tests were best suited to the formality questions.

To aggregate orientations so as to produce more indicative results and larger groups of responses, I also grouped responses for some statistical tests. Two kinds of groupings were done:

• item grouping: Responses to a historic on form 1 were grouped with responses to an historic on form 2, and responses to an historical on form 1 were grouped with responses to a historical on form 2; this does not filter for other possible reasons for assent or dissent.
• four-way grouping: Respondents were classified according to whether they (1) said yes to an and no to a on the two historic/al questions on their form; (2) said no to both; (3) said yes to both; or (4) said no to an and yes to a on the two historic(al) questions on their form. Respondents who said uncertain to either question were put in group 0 and excluded. In the final calculations, the numbers in group 2 were too low to allow reliable calculations, and so they were excluded.

Findings

Basic results

The total responses for the a/an items (including an hotel) were as follows (the number in parentheses after each indicates which form it was on):

		a historic (form 1)	an historic (form 2)	a historical (form 2)	an historical (form 1)	an hotel (form 1)	an hotel (form 2)
correctness	incorrect	62	30	45	36	79	65
	correct	32	79	59	60	27	31
	uncertain	9	1	6	7	4	7
formality	mean	3.214	3.773	3.164	3.379	2.903	2.664
	variance	1.268	0.801	0.799	0.845	0.971	1.069

Overall, it is clear that an historic is preferred to a historic by a clear margin (nearly two to one), and that an historical is preferred to a historical by a small margin, but an hotel is considered incorrect by a margin well over two to one. We see also that the an variants of historic and historical are considered more formal; however, a Student’s t-test reveals that the difference in perceived formality between a and an is statistically significant for historic (at p<0.001) but does not reach significance for historical (p=0.085). Likewise, there is a significant difference in perceived formality between an historic and an historical (p=0.002), but not between a historic and a historical (p=0.72). We also find that the an variant is seen as comparatively informal for hotel, significantly so (p≤0.001) for all variants.

Correctness choice relationships

There were significant relationships between individuals’ choices on the a/an items for three of the six possible pairings of responses (three pairings for each respondent: for form 1 respondents, a historic with an historical and an hotel, and an historical with an hotel; for form 2 respondents, the same with a and an reversed for historic and historical). There was no relationship in choice of correctness between a historic and an hotel. Likewise, there was no statistically significant relationship in choice of correctness between a historical and an hotel. The relationship between an historical and an hotel also failed to reach significance. However, the relationship between an historic and an hotel was significant, as were the relationships between a historic and an historical and between an historic and a historical. For the most part, these actual results differed from the predicted by about a 3:2 or 2:3 margin; all three of the relationships were significant at p<0.001 in chi-square tests.

		an hotel
		incorrect	correct
an historic	incorrect	29	1
	correct	49	26
predicted values
		22.29	7.71
		55.71	19.29
		p=0.0009

		an historical
		incorrect	correct
a historic	incorrect	14	45
	correct	19	11
predicted values
		21.88	37.12
		11.12	18.88
		p=0.0003

		a historical
		incorrect	correct
an historic	incorrect	2	42
	correct	27	32
predicted values
		12.39	31.61
		16.61	42.39
		p=0.000004

Correctness and formality

For formality relationships, Student’s t-tests were used, since usable means and variances could be calculated from the 5-value scale. Chi-square analyses would have been less reliable due to the low number of data points in some of the cells.

	a historic		an historic		a historical		an historical
	incorr	corr	incorr	corr	incorr	corr	incorr	corr
mean formality	3.016	3.688	3.300	3.975	3.044	3.288	2.958	3.500
variance	1.229	1.190	1.045	0.563	0.907	0.657	0.923	0.763
	p=0.007		p=0.002		p=0.172		p=0.070

	a historic		an historic		a historical		an historical
four-way group	1 (an)	4 (a)	1 (an)	4 (a)	1 (an)	4 (a)	1 (an)	4 (a)
mean formality	3.089	3.684	4.048	3.370	3.119	3.444	3.489	3.211
variance	1.083	1.117	0.632	1.088	0.839	0.795	0.665	0.509
	p=0.046		p=0.006		p=0.149		p=0.180

As we see, for historic, those who preferred an found it significantly more formal than those who preferred a, but, while there was a similar effect for historical, it did not reach significance at the p<0.05 level. When we look within the four-way groups at ratings of different a/an versions, we see small but statistically non-significant differences for most pairings; the notable exception is the difference in ratings of a and an for historic in group 1, which is a difference of almost a full point in average, significant at p=0.00006. Group 4 did not produce a statistically significant difference for this pairing. Differences in group 3 (both correct) averages were comparatively small and did not approach statistical significance:

	a historic	an historic	a historical	an historical
mean formality	3.636	3.844	3.156	3.636
variance	1.655	0.459	0.523	1.455
	p=0.574		p=0.236

As well, the hypothesis that an would be found more formal even among those who thought it incorrect did not hold up.

Relationships with other items

It was hypothesized that there would be a relationship between preferring an and preferring common prescriptive norms such as proscriptions on splitting infinitives, ending sentences with prepositions, and using hopefully as a sentence adverb. However, no such effects were found. There was, however, a small but significant relationship between a historic and She gave it to John and I:

		John and I
		incorrect	correct
a historic	incorrect	55	7
	correct	23	9
predicted values
		51.45	10.55
		26.55	5.45
		p=0.04

We see a slight tendency for preference for a to go with preference for John and I, and for preference against the one to go with preference against the other. There was an apparent relationship of similar degree and implication between an historic and She gave it to John and me, but it failed to reach p<0.05 significance:

		John and me
		incorrect	correct
an historic	incorrect	12	16
	correct	19	58
predicted values
		8.27	19.73
		22.73	54.27
		p=0.07

The other pairings were not available due to the distribution of the items on the forms. No other significant relationships were discerned.

Demographic effects

Since correlation calculations were not possible due to the non-scalar choice in correctness judgement, respondents were put into four age groups, which were determined on the basis of age distribution among the respondents: 0–34; 35–51; 52–64; 65+. Other groupings were tried and did not produce clearer or significantly different results. Although age-related effects did not reach significance within individual questions, when grouping was applied (as described in the Methods section, above), significant age-related effects were found through both grouping techniques.

Item grouping:

		historic
		an	a
age group	1	25	23
	2	36	15
	3	43	11
	4	37	13
predicted values
		33.34	14.66
		35.42	15.58
		37.51	16.49
		34.73	15.27
		p=0.02

		historical
		an	a
age group	1	21	26
	2	20	28
	3	36	17
	4	28	24
predicted values
		24.675	22.325
		25.2	22.8
		27.825	25.175
		27.3	24.7
		p=0.04

Four-way grouping:

		a/an group
		1 (an)	3 (both)	4 (a)
age group	1	16	8	17
	2	16	16	10
	3	31	8	8
	4	24	11	11
predicted values
		20.27	10.02	10.72
		20.76	10.26	10.98
		23.23	11.48	12.28
		22.74	11.24	12.02
		p=0.02

The results show a clear higher-than-predicted preference for an among those in age group 3, 52–64 years, and a clear higher-than-predicted preference for a among those in age group 1, 0–34. Age group 2 is slightly more varied, tending to prefer a historical by a small margin but not differing notably from predicted values for a/an historic; however, this age group is also more likely to choose both as correct rather than to side with a or an exclusively. Age group 4, 65+, also does not differ notably from predicted values.

There was no statistically significant age-related effect with an hotel.

There were no statistically significant age-related effects for formality in any of the a/an questions, regardless of the statistical means used (chi-square, Student’s t-test, or correlation).

There were no significant effects for any of the three a/an choices for level of education.

For place of education, Canada and the US were grouped together (due to the small number of US-educated respondents and the commonality between the two countries in pronunciation of [h] in the words in question) and Britain and “elsewhere” were grouped together (due to the small number of responses and the general tendency towards British-style dialects in other countries). Respondents who had been educated in both North America and Britain or elsewhere were put in the Britain/elsewhere group by reason of their at least having been exposed to a British approach, which those exclusively educated in North America would not be expected to have. There were no significant effects for any of the three a/an choices for place of education. Note that this was the case even for an hotel, which might have been expected to be an indicator of British education.

Respondents were asked whether English was their first language; as only 13 of 213 respondents said it was not (3 on form 1 and 10 on form 2), it was not possible to use this information to draw inferences.

There was no significant effect for gender (sex) for any of the a/an choices.

Analysis

The clear relationship in correctness judgements between a and an on the historic(al) items shows that the judgements on those items was in fact focused on that specific variable and not on some other detail of the stimulus sentence. We see that there are three camps: those who consider only an correct (this is the largest group by a fair measure), those who consider only a correct, and those who accept both as correct. There were very few who considered neither correct. Thus the a/an grouping is valid and useful. The fact that I was only asking whether a specific sentence was correct or not led to different results than I would have gotten from recording the respondents’ own actual usage; we can see that nearly as many of them (43) fell into a/an group 3 (both are correct) as into a/an group 4 (a only; 46). But there were nonetheless nearly as many (87) in a/an group 1 (an only) as in the other two groups combined.

Although it would have been useful to know with some certainty which variant the respondent actually used, the difficulty with simply asking the respondents this is that the direct focus on the question might have had too much of a skewing effect on the results. In a study with more time and resources, eliciting a spoken sentence that included a or an historic before proceeding to the questionnaire would have been of use. If we take correctness judgement on an hotel to be an at least modestly reliable indicator of actual pronunciation, then we can postulate that there may have been an influence of pronunciation, but not an exceptionally strong one; only an historic showed a significant relationship with an hotel, and the difference was on the order of 6 to 7 respondents. As well, there was no significant effect for place of education with any of the items. It would thus seem that judgements on the correct article to use were split even among those who always pronounce [h] in these words. And the absence of a relationship between place of education and a/an correctness preference indicates that the split in preference is current in both regions, which matches what has been seen in the web searches (in the Background section, above).

One factor that should not be ignored is the possibility of a person considering an correct and a incorrect even though he or she knowingly uses a as a matter of course. Many people are used to speaking “incorrect” English much of the time – English which, to their knowledge, is not formally “correct” but is nonetheless the version they prefer to speak because it is the language of their peers and they are more comfortable with it. This possibility is bolstered by the fact that a few of the respondents commented to me after doing the survey that they were curious as to how many they “got right.” For many Canadian speakers, it is quite possible that an historic is a postvernacular usage. As Preston (2004) explains, “adult learners of their own language encounter syntactic (and other) characteristics that they learn in no substantially different way than the second- or foreign-language learner learns things…, and I have no reason to assume that they end up embedded in the underlying grammar in any significantly different way.” The possible disjunction between correctness judgement and actual usage is of considerable interest and would merit a subsequent study focused on it.

Perceived formality had an important relationship with correctness judgement. A/an group 1, those who considered an correct and a incorrect, considered an historic significantly more formal than a historic, while group 4, those who considered a correct and an incorrect, did not have a statistically significant difference in rating. The hypothesis that even those who found an historic incorrect would consider it more formal did not hold up. This suggests that for those who prefer an, the choice is a matter of formal correctness, and formality is important, whereas for those who prefer a, formality does not enter the issue in a significant manner. This may be seen to have a connection to the perception of an-preference as the territory of sherry-sipping, bowtie-wearing snobs: that is, it has a connection to an ideology of formal correctness. Note, however, that this is only the case for historic, not for historical – the formality focus is strongly on that specific word.

This strong formality effect for historic is reasonable, given that historic is more given to formal and momentous usages. The ITP Nelson Canadian Dictionary of the English Language (1998, 644) gives this usage note, which well characterizes the common distinction in usage: “Historic and historical are differentiated in usage, though their senses overlap. Historic refers to what is important in history: the historic first voyage to outer space. It is also used of what is famous or interesting because of its association with persons or events in history: a historic house. Historical refers to whatever existed in the past, whether regarded as important or not: a historical character. Historical refers also to anything concerned with history or the study of the past: a historical novel. The words are often used interchangeably: historic times or historical times.” One might be led to posit the existence of two perceived versions of historic, one more formal taking an, the other less formal taking a. This is supported by the facts that an historic was seen overall as significantly more formal than a historic and significantly more formal than an historical, while a historic was not seen as significantly more formal overall than a historical. However, some of this effect will certainly be due to choice among those who consider only one of the versions correct (and who therefore would not have an impression of two equally valid versions with differing formality). When we look only at the “both a and an” group (3), an historic was not rated significantly more formal than a historic or an historical.

We thus have something of an account for the special persistence of an historic and a possible suggestion of two versions of historic that may exist for many users: one that takes a and is neither formal nor properly “correct,” and the other that takes an and is formal and properly “correct.” The correctness judgements for historical may have been pulled along by the judgements for historic; if there had been a group of respondents large enough to allow for four forms, so that historical could have been tested without historic on the same page as a possible influence, there may have been different results. As it was, the correctness judgements for historical were not as strong as for historic. Nonetheless, we can see from the surveys of news outlets and websites in general that historical is still actively given an by many users independently of historic, so the ideology of formality, while it may have some effect, is clearly not the only determining factor; the history of this usage also plays a part. But it is not unreasonable to hypothesize that an historic, with its connections to an ideology of formality, is the anchor of this conservatism, and that an historical and, to a lesser degree, a few others are being maintained to a fair degree by the influence of an historic.

We cannot, however, posit a larger set of prescriptive norms that form a coherent unified standard including an historic. The hypothesized relationships between a/an historic(al) judgements and judgements on items that manifested characteristics subject to strong, often ideologically based prescriptive preferences simply did not manifest. This does not mean that there is no ideological basis, of course, only that such basis as may exist operates independently from other ideological orientations that may motivate the other judgements.

The sole important demographic effect identified was age-group related. There are different possible reasons for this effect. One possibility is that it is indicative of a sea-change in attitudes towards English and its teaching. The advent of “whole language teaching” in the late 1960s and 1970s, and its ascendancy in the 1980s and thereafter, may be of debatable merit overall, but it is clear that it de-emphasized rote learning and dogmatic approaches to English usage. We might note that those in age group 1 (34 and under) would have graduated from high school starting in about 1990, and thus would have begun their schooling in the late 1970s or later, just when “whole language” was reaching its ascendancy. Those in group 3 (52–64), on the other hand, would have finished high school no later than the early 1970s and would have done their schooling almost entirely in the 1950s and 1960s. As to those in group 2 (35–51), they were in school during the time that “whole language” was just coming and rote learning was on its way out. And it may be that a vogue for an historic passed through North American (and perhaps British) usage in the 1960s and 1970s, as Bolinger (1975) suggests; this is worth a further study.

On the other hand, it is possible that this age difference reflects standard sociological age-related norms of usage at least as much as it does any real change passing through the language; as Eckert (1997) notes, increased conservatism is an important linguistic change for those entering the adult phase of life, and a relaxation of conservatism is thought to be characteristic of those who have reached retirement age. This hypothesis is supported by the lack of a clear effect for group 4 (65 and over). Further study would be required to separate out the effect of increased conservatism in mid-life and to determine whether, in fact, a permanent change is passing through the language.

One possible effect on the difference between the generations may be the comparative absence of postvernacular learning of an with historic among the younger generation; another, converse, possibility is that the older generation tended to learn an in school as a more integral part of their language, and it is the younger generation, exposed to it in a more desultory fashion, who acquire it postvernacularly if at all. However, we need to take note of the absence of a relationship between age and formality judgement. While age and formality judgement both have relationships with correctness judgement, they operate independently. Thus, it does not appear that members of age group 3 learned that an historic is formal while members of age group 1 did not. It seems, rather, that they merely learned whether it is correct, and the formality perception is a separate (perhaps postvernacular) learning that we might imagine derives from the real-life contexts in which the usage has been encountered and from a given person’s own disposition towards the ideology of formality and correctness.

The predominance of a in style guide recommendations might be taken as some indication that there is indeed a permanent change in this usage gradually making its way through the population. But we have observed, in the Background section above, that a was already widely recommended over an by style guides before any of the respondents to my survey were born (the oldest respondent was 88 years old, meaning she was born in 1918 or 1919), and certainly long before most of them were in school. So why would they not have learned what usage manuals counsel? One likely reason is that, in general, they did not read the manuals and were not taught from them. A survey of school texts from the various decades of the 20^th century and from different parts of the English-speaking world would be informative with regard to what people were, in fact, taught and when, but such a survey is far beyond the means available for this study. An even better, but even less possible, survey would be of the attitudes of the English teachers themselves from the course of the 20^th century; personal experience tells me that many people will cleave barnacle-like to the dogmas propounded by their high school English teachers even in the face of contradiction by what one would think would be greater authority. On the other hand, many others will forget what they were taught and will conform their usage to what they are used to seeing and hearing. The question then remaining is, given that a historic definitely outweighs an historic in current Canadian usage and has been the prescribed standard in most quarters for nearly a century, how are respondents coming to prefer an historic? The formality connection suggests an answer to this: an historic is sometimes seen in formal contexts, and the presence of the marked an is inferred to be correct precisely because it is exceptional – else why would the expected a not be used? – and from this, to the extent that the person values the linguistic ideology of formality and correctness, a judgement of formality and correctness is formed, even in the face of a majority of usages of a historic, which are discounted as common but incorrect.

Conclusion

An historic and, to a lesser degree, an historical present us with an example of persistence of an exception to a very well-established rule of usage. Although the rule in English is that the choice of indefinite article is determined by the initial sound of the immediately following word, some users overtly discard this rule in this one instance, and others claim that this instance is a special case where a consonant becomes no longer a consonant. The history of this usage gives us a good sense of its origins, but we would expect the usage to have disappeared almost entirely by the present time as have most other similar usages. Instead, it persists. The survey of 213 mostly North American English speakers sheds some light on the phenomenon: there is special influence from the perceived formality of this word, and there is also an age-related effect. All age groups preferred an historic to a historic, but the youngest group somewhat less so and group 3 most so; only age group 3 preferred an historical to a historical by a significant margin. The age-related effect may be a static one whereby users move from a phase of a-preference in youth through a phase of an-preference in middle adulthood and back towards a in their older years, or it may be a real change that is gradually moving up through the populace, or both effects may be in operation. The surest way to determine which is the case would be to continue to survey users every several years. It does seem likely that styles of English education will have some bearing on the matter; perhaps the best way to guarantee solid dominance for a would be for English teaching to return to a more prescriptive style, specifically one in which a historic and a historical are taught as correct and an historic and an historical as nothing but a pair of mumpsimuses. Failing that, the current state of affairs, in which usage is learned much more through folk learning and inference, could allow the current common division to persist for a long time yet, with many, perhaps even a majority, judging an to be the correct version even as a may be more often spoken – and by far more commonly recommended in style guides and usage manuals.

References

Bolinger, Dwight (1975). “Are You a Sincere H-Dropper?” American Speech 50(3/4): 313–315.

Bollard, J. K. (1979). “A or An?” American Speech 54(2): 102–107.

Burchfield, R.W., editor (1996). The New Fowler’s Modern English Usage. Oxford: Clarendon Press.

Byington, Steven T. (1929). “‘A’ and ‘An’ before ‘H’.” American Speech 5(1): 82–83.

City Comforts, temporarily known as Viaduct, The Blog (2006). “Department of Correct Usage.” Posting by “Chris Burd.” citycomfortsblog.typepad.com/cities/2006/05/department_of_c.html, accessed March 20, 2007.

Eckert, Penelope (1997). “Age as a Sociolinguistic Variable.” In The Handbook of Sociolinguistics, ed. Florian Coulmas, 151–167. Malden MA: Blackwell.

Editors’ Association of Canada (2000). Editing Canadian English. Toronto: Macfarlane Walter & Ross.

Fee, Margery, and McAlpine, Janice (1997). Guide to Canadian English Usage. Toronto: Oxford University Press.

Feipel, Louis N. (1929). “‘A’ and ‘An’ before ‘H’ and Certain Vowels.” American Speech 4(6): 442–454.

Hillhouse, J.T. (1928). “A or An?” Modern Language Notes 43(2): 98–101.

Irvine, Judith T. (1989). “When Talk Isn’t Cheap: Language and Political Economy.” American Ethnologist 16:248–267.

ITP Nelson Canadian Dictionary of the English Language (1998). Toronto: ITP Nelson.

Lane, Ralph H. (1949). “Modern ‘Historic’.” American Speech 24(3): 181–188.

Liberman, Mark (2004). “Hung Like a Hero.” itre.cis.upenn.edu/~myl/languagelog/archives/000463.html, accessed March 18, 2007.

Kenyon, J.S. (1929). “‘A’ and ‘An’ before ‘H’.” American Speech 5(1): 84–85.

Kroch, Anthony, and Small, Cathy (1978). “Grammatical Ideology and Its Effect on Speech.” In Linguistic Variation: Models and Methods, ed. David Sankoff, 45–55. New York: Academic Press.

Milroy, Lesley (2004). “Language Ideologies and Linguistic Change.” In Sociolinguistic Variation: Critical Reflections, ed. Carmen Fought, 161–177. Oxford: Oxford University Press.

Opinion L.A. (2007). “Non-genuine article makes reader sic.” Postings by “Tim Cavanaugh,” “Rob McMillin,” and “Brady Westwater.” opinion.latimes.com/opinionla/2007/01/nongenuine_arti.html, accessed March 20, 2007.

Palmer, Francis L. (1929). “‘A’ and ‘An’ before ‘H’.” American Speech 5(1): 83–84.

Preston, Dennis R. (2004). “Three Kinds of Sociolinguistics: A Psycholinguistic Perspective.” In Sociolinguistic Variation: Critical Reflections, ed. Carmen Fought, 140–158. Oxford: Oxford University Press.

Silverstein, Michael (1979). “Language Structure and Linguistic Ideology.” In The Elements: A Para Session on Linguistic Units and Levels, ed. Paul R. Clyne, William F. Hanks, and Carol R. Hofbauer, 193–247. Chicago: Chicago Linguistics Society.

Tasko, Patti, editor (2005). The Canadian Press Caps and Spelling, 17th edition. Toronto: The Canadian Press.

Urban Dictionary (2004). “an historic.” Posting by user “Kung-Fu Jesus.” http://www.urbandictionary.com/define.php?term=an+historic, accessed March 20, 2007.

Walsh, Bill (2006). “More on ‘A’ vs. ‘An’.” http://www.theslot.com/a-an.html, accessed March 18, 2007.

Woolard, Kathryn A., and Schieffelin, Bambi B. (1994). “Language Ideology.” Annual Review of Anthropology 23: 55–82.

Yahoo Answers (2006). “What is correct English: ‘a historic’ vs. ‘an historic’… What’s the difference?” Postings by users “James” and “DanielSay.” answers.yahoo.com/question/index?qid=1006020106262, accessed March 20, 2007.

Yahoo Answers U.K. & Ireland (2007). “A historic or an historic event? At school I was taught the article ‘a’ only precedes a consonant initial word.” Posting by “erindehart1”. uk.answers.yahoo.com/question/index?qid=20070313134721AA31uVK, accessed March 20, 2007.

Appendix 1: Survey forms

When respondents went to http://www.harbeck.ca/ling/ as directed, they were presented with this page:

The Formality and Correctness Survey

Hello, and thank you for coming to the formality and correctness survey. This is a brief survey (20 study questions and eight demographic information questions) to gather information on the perceived formality and correctness of certain English usages. It is being done for an assignment for LING 3650, Sociolinguistics, at Glendon College, which is part of York University. It will most likely take you less than two minutes to complete.

This is an anonymous survey, so you are on your honour to complete it only once – please do not come back and do it again, as that will skew the data. But please do ask friends and family to complete it as well. All data gathered for this survey by March 7, 2007, will be included in the analysis. I will have no way of connecting a specific set of data with a specific respondent, because the form does not collect your name, address, IP address, or any other information sufficient to identify you personally. Clicking on “Send” when you have completed the survey indicates your agreement to participate in this study and your agreement with the terms and manner of its conduct.

If you would like to read the results and analysis of the survey, please email me, James Harbeck, at james@harbeck.ca, and I will send them to you once the survey and assignment are complete.

To begin, please click on the last digit of your postal code. This will allow me to sort the results.

0

1

2

3

4

5

6

7

8

9

Clicking on their postal code took them to one of two forms. Form 1 was used for people with postal codes 0–4. Form 2 was used for people with postal codes 5–9. The questions on the forms were not numbered on the page the users saw; however, the responses relayed to me by the form were numbered (or, on form 2, lettered). These numbers, and other accessory information not visible to the users, are included below in square brackets. The responses were entered by clicking on “radio buttons” for the formality and correctness ratings and most of the other points of input, an input field for age, and checkboxes for country of education. To save space, I will not reproduce the formality and correctness input field after each item below; it was formatted in the following manner (the o’s represent radio buttons):

informal			formal		|	correct?
1	2	3	4	5	|	no	yes	uncertain
o	o	o	o	o	|	o	o	o

The forms were identical except for the 20 stimulus items; thus, the form is presented once below, with the stimulus items side-by-side in table format.

In the responses sent by the form, the formality ratings were part a of the question and the correctness ratings were part b; therefore, if a person rated item 5 as a 4 on formality and a “yes” on correctness, I would receive “5a: 4” and “5b: y”.

The Formality and Correctness Survey

Please rate the style and tone of each of the following sentences on a scale of 1 to 5, where 1 is informal and 5 is formal, and please indicate whether you consider the usage correct – whether the sentence is “good English” (click on “uncertain” if you’re not sure). Please do this for all of the phrases – don’t skip any.

[form 1]	[form 2]
[1] We want to aggressively pursue this opportunity.	[A] We wish to aggressively pursue this opportunity.
[2] There are a lot of reasons to do so.	[B] There is a lot of reasons to do so.
[3] This is something which we must address.	[C] This is something that we must address.
[4] I think it’s fun.	[D] I think it’s fun.
[5] This is a historic occasion.	[E] This is an historic occasion.
[6] Program director, Margaret Wilson says that eleven courses will be offered.	[F] Program director Margaret Wilson says that eleven courses will be offered.
[7] The move is misguided and, more important, it may do positive harm.	[G] The move is misguided and, more importantly, it may do positive harm.
[8] You’ll love this atlas, published by the National Geographic Society.	[H] Published by the National Geographic Society, you’ll love this atlas.
[9] The job was done by the director.	[I] The director did the job.
[10] Hopefully, we will not need to repeat this exercise.	[J] Hopefully, we will not need to repeat this exercise.
[11] I’m not going to do it.	[K] I ain’t going to do it.
[12] She gave it to John and I.	[L] She gave it to John and me.
[13] They conducted an historical survey.	[M] They conducted a historical survey.
[14] I’m glad you came.	[N] I’m glad that you came.
[15] You want me to do what?	[O] You want me to do what?
[16] We thought it was done; however, it was not.	[P] We thought it was done, however it was not.
[17] These kind are not so good.	[Q] This kind are not so good.
[18] There was an hotel on the other side of the river.	[R] There was an hotel on the other side of the river.
[19] That was the place I had heard of.	[S] That was the place of which I had heard.
[20] Thank you for doing this.	[T] Thanks for doing this.

Thank you! Now please give me some demographic information, as it may be relevant to differences in perception. (Remember, this is all anonymous.)

[age] How old are you?

[input field] years

[sex] Are you [button] male or [button] female?

[efl] Did you grow up speaking English?

[button] yes

[button] no

[edu] What is the highest level of education you have completed?

[button] less than high school

[button] high school

[button] some university or college

[button] bachelor’s degree

[button] graduate or professional degree

Please put a checkmark beside the country or countries where you attended primary and/or secondary school:

[Can] Canada

[USA] USA

[Eng] England

[els] elsewhere

Once you’ve answered all the questions, please click “Send.”

[Send button]

If a respondent failed to answer a question, the form would not allow him or her to submit it until a value had been entered in the field. Once the respondent clicked on “Send,” a page appeared listing the response values that were sent to me. This was a feature of the software that received the form and sent me the email; it was not something I was able to change.