The problem with recognize and its variant pronunciations

Investigating the phonological patterns, especially the stress patterns, of verbs ending in -ize such as finalize, constitutionalize, etc, the word recognize has attracted my attention. One would not generally attach too much attention to this word for its phonology: it seems to be a run-of-the-mill case of stressing the third-last (antepenultimate) syllable of a non-transparent derivation by -ize. For instance, Nádasdy (Reference Nádasdy2006: 222) treats -ize as a basically neutral (strong) suffix, that is one that is not supposed to interfere with stress-patterns and otherwise of the stem to which it is attached. Following established analyses, he divides -ize words into two categories, though: those that are derived from a free stem ('character > 'characterize, 'final > 'finalize), where stress (indicated by the ' mark) does not shift in the derived verb, and those whose stem is non-transparent ('recognize, 'categorize), and where stress tends to be furthest away from the suffix itself. The fact that recognize has a non-transparent derivation means that there is no free English word *recogn. Ginésy (Reference Ginésy2004: 126) analyzes recognize as morphologically having a double prefix, re- and co-, which reduces the stem to Latinate -gn-, which is always bound in English. Whether his etymological analysis is warranted for the contemporary morphology of recognize is at least disputable today, but he correctly claims that 'recognize, with stress on the initial syllable, behaves like a non-transparent derivation so it receives antepenultimate stressing. In other words, the most wide-spread pronunciation of recognize, with initial stress, is generally unproblematic in the literature: it is a case of non-transparent derivation by -ize with antepenultimate stressing.

Curiously, however, Wells (Reference Wells2008) lists two further possible pronunciations for recognize. One is a ‘British English but non-Received Pronunciation’ variant where stress is final: recog'nize /rekəɡ'naɪz/. Even more interestingly, he points out a pronunciation ‘generally considered incorrect’: /'rekənaɪz/, that is without /ɡ/. This /ɡ/-less pronunciation is listed as a regular variant in the Longman Dictionary of Contemporary English (LDOCE, 2009). It is mainly this latter variant that deserves special attention. From an orthographic point of view, one could be tempted to jump to the hasty conclusion that <g> is here a silent letter just like the <g> in sign, malign, paradigm or phlegm and a few others, all of them borrowed from French, where the <g> before a nasal <m> or <n> is mute in English. This /ɡ/-less pronunciation of recognize is worth discussing at some length because in fact it does not conform to established patterns of silent <g> and also because there does seem to be a phonological explanation, rather than one based on etymology, for the relative currency of this /ɡ/-less variant. The tentative explanation proposed in this paper has to do with the suffix -ize itself, and the form can be analyzed as a special case of non-transparent derivation.

Patterns of silent <g> – recognize does not fit them

The orthographic rule for silent <g> seems not to be convincing as accounting for the /ɡ/-less variant of recognize, because a silent <g> is never found in the position where recognize /'rekənaɪz/ has it. It is to be noted here that in initial <gn > , every <g> is silent (gnaw, gnostic, etc) – such examples are excluded as irrelevant from the present discussion altogether, because the silent <g> examined here occurs among the letters in coda (final) rather than onset (initial) syllabic positions. A pre-nasal silent <g> occurs among the letters in the coda of a stressed final syllable or, more rarely, of an unstressed final syllable (this syllable having either a full vowel or a reduced vowel). To collapse these environments: silent <g> occurs among the letters in the coda of final syllables. Otherwise the <g> is regularly pronounced before a nasal. Examples are shown in (1a–b) below for silent <g>, while (1c–d) show examples of pronounced <g> before a nasal.

The patterns look somewhat involved but are not inextricably complex. Etymological pre-nasal <g> is silent in English when it comes in the final (or only) syllable of a content word, irrespective of whether or not it is stressed in a polysyllabic word. Data in Table 1 columns (a) and (b) cite a few examples. Most frequently, the vowel of this syllable is a stressed or unstressed diphthong, /eɪ/ or /aɪ/, although phlegm,Footnote ¹apophthegm and diaphragm, with short stressed vowels, and sovereign, foreign, with a reduced vowel, behave in the same way: the <g> is silent. The words in column (b) are somewhat heterogeneous: paradigm and diaphragm have a full vowel, and for this reason they are given some prosodic prominence in some analyses, while foreign (modern French feminine foraine) and sovereign (modern French feminine souveraine) have an unetymological <g>. This extra <g> is possibly due to the digraph spelling of the final syllable of these words, which used to rhyme with reign and feign, where the <g> is etymological. These two words are cases of orthographic analogy, underlining the contemporary observation that <g> is silent when it comes in the final syllable. The important point to make here about recognize is that the /ɡ/-less pronunciation /'rekənaɪz/ does not fit either of the ‘silent patterns’ since its putative silent <g> is not final.Footnote ²

Table 1: < g> before a nasal letter in English and stress-patterns

V = vowel, C = consonant, () = optional, # = word boundary marker, ' = primary stress

(examples taken from Nádasdy, Reference Nádasdy2006: 82, Sobkowiak, Reference Sobkowiak2004: 120, and Wells, Reference Wells2008)

On the other hand, the most widespread variant, /'rekəgnaɪz/, is also problematic with respect to silent letters. As columns (c–d) of Table 1 show, orthographic pre-nasal <g> is regularly pronounced when it comes in a non-final syllable, where it is either in the coda of a stressed syllable followed by an unstressed syllable or where it is in the coda of an unstressed syllable followed by a stressed syllable. In other words, orthographic pre-nasal <g> is pronounced when it is the coda at the boundary of a stressed and an unstressed syllable. It does not matter in what order the stressed and unstressed syllables follow each other in the word. Numerous examples in (c–d) have a weak suffix: -tion, -ic. Incidentally, this latter phonological context accounts neatly for the non-RP final-stressed pronunciation /rekəg'naɪz/ listed by Wells (Reference Wells2008): the stressed syllable -'nize includes the suffix -ize. (The final stressing of this variant also implies that -ize, when it attracts stress on itself, is exceptionally a weak suffix rather than a strong suffix as generally assumed – but this issue will not be pursued any further here.) It should also be noted in this context that recog'nition has the primary stress on this syllable, which is perfectly regular, and that recognizable can be pronounced with primary stress on -'nizable.

As to the two remaining variant pronunciations of recognize, it can be observed that no single pattern fits either variant of recognize. On the one hand, /'rekəgnaɪz/, with initial stress and /ɡ/, fits neither the ‘silent patterns’ nor the ‘pronounced patterns’, since its <g> is not final to be silent and not at the boundary of a stressed and an unstressed syllable to be pronounced (<g> is at the boundary of two unstressed syllables). On the other hand, /'rekənaɪz/, with initial stress but no /ɡ/, also does not fit the ‘silent patterns’ since <g> is not final, and it does not fit the ‘pronounced patterns’ since <g> is not at the boundary of a stressed and an unstressed syllable. The closest analogy for this stress pattern is 'stalagmite /'stæləɡmaɪt/, which has /ɡ/ in exactly the same phonotactic position – but there are no recorded variants for this word. The verb impregnate /'ɪmpreɡneɪt/ is also similar for its stress-pattern and even has a verbal suffix, -ate, but it does not have a reduced vowel before /ɡ/ – and again, no variants are recorded. The fact that these words also have a pronounced /ɡ/ show that the rule aims at silent letters: if there is no reason for the <g> to be silent, it will be pronounced. In this section it has been pointed out that the /ɡ/-less variant of recognize does not fit established patterns of silent <g>, and it has also been noted that in fact neither the /ɡ/-ful nor the /ɡ/-less variant have anything to do with silent letters. If recognize can be pronounced without /ɡ/ (unlike stalagmite and impregnate), it must have a very different reason from the orthographic rule of silent <g>.Footnote ⁶

Before moving on to discussing why there is a /ɡ/-less variant for recognize, it should perhaps be added for clarity's sake that, from a phonological point of view, there is no silent /ɡ/ in any of the above words: the mismatch between the pronunciation and the spelling of these words is totally irrelevant. There is no phonological sense in saying that sign /'saɪn/ has a silent /ɡ/ while sine /'saɪn/ does not. The two are homophones today. It is only their morphological alternations that betray their different etymologies: sign happens to have relatives like signal or signature, both with /ɡ/, while sine has sinusoid, which lacks /ɡ/. But sign and sine do not have silent sounds when pronounced, of course.

A suggested (morpho)phonological explanation why /'rekənaız/ is possible

The reason why it is important to look for a contemporary phonological explanation for /'rekənaɪz/ is because recognize is not obviously perceived as a French word in contemporary English. Moreover, it is definitely not used in contemporary French with any comparable frequency, if it can be said to still exist in the first place. According to the OED, recognize is first attested in 1388/89 in English, so it is not a recent borrowing. It is also known that the /ɡ/-less variant has been around at least since 1791, and probably much earlier, because Walker (Reference Walker1791: 45)Footnote ⁷ makes a comment on its usage: ‘Some affected speakers, either ignorant of the rules for pronouncing English, or over-complaisant to the French, pronounce physiognomy, cognizance and recognizance, without the n;Footnote ⁸ but this is a gross violation of the first principles of spelling.’ This would suggest that affected speakers of English, trying to imitate French /ɲ/ or /ɡ/-less pronunciations, had taken up recognize without /ɡ/: it is then a later variant, and the /ɡ/-ful variant can legitimately be recognized as primary. Interestingly, it is these three words that still show variation between the presence or absence of /ɡ/: Wells (Reference Wells2008) has a /ɡ/-less variant for both physiognomy and cognizance. It is possible that Walker had tapped into some contemporary English phenomenon, which he interpreted to be affected speech. For instance, modern French has physionomie, without <g> in the spelling or /ɲ/ in the pronunciation, and the French agent noun physionomiste is attested since the sixteenth century and seems to have always been spelt without <g>. For physiognomy and its derivatives at least, it seems safe to say that English did not borrow /ɡ/-ful versions from French in the first place, but has inserted one based on Latinate etymology just as in dou bt and de bt, where the <b> does not come from French. This would make the /ɡ/-ful forms spelling pronunciations. Furthermore, the cognate forms of recognize/cognize seem to have been obsolete by this time in French. It is hard to see how they could serve for any model, then – the modern form (re)connaître could not influence recognize. The point to be made here is that although silent <g> heavily overlaps with real or imaginary French origin, the contemporary patterns of silent <g> do not cover the case of recognize (or physiognomy for that matter). Consequently, it must have some phonological importance that this group of words, but not stalagmite or impregnate, show variant pronunciations. While the etymological explanation has its merits in accounting for why a /ɡ/-less variant emerged in the first place, it is somewhat simplistic, because there seem to be reasons to think that recognize has undergone some morphophonological reinterpretation in its history. In other words, modern English recognize must have better reasons for keeping a /ɡ/-less variant than merely claiming French origin.

There is an interesting morphophonological observation concerning recognize. It is known that the addition of strong-boundary suffixes such as agentive -er, -ing or -ment does not interfere with the pattern of silent <g>: designer, signing, alignment – all without /ɡ/. The suffix -ize, as pointed out above, is generally considered a strong or neutral suffix: therefore it should not interfere with silent <g>. As a matter of fact, one could propose that the pronunciation of recognize without /ɡ/ is a good indicator that -ize is indeed a strong-boundary suffix: the verb clearly comes from /'rekən/ # /aɪz/ > /'rekənaɪz/. The form /'rekən/ is incidentally an attested free-standing phonological form in English, spelt <reckon>. This word is etymologically distinct from recognize. From a phonological point of view, however, the form /'rekənaɪz/ does look as if derived from *reckon > reckonize. The morphological problem is that reckon is not used as a noun or adjective in contemporary English, not even in idioms. This is problematic, since -ize attaches to noun and adjective stems, not to verbs. It can be said that this is a non-transparent derivation, in the sense that only the phonological form of reckon is needed for the derivation. Nevertheless, even semantically the derivation from reckon 'to calculate an amount, to guess a number without calculating it exactly' seems to fit in with the general semantics of -ize suffixation: recognize ‘to accept or admit that something is true'. At any rate, the two meanings are probably close enough to have a potential interference even through folk etymology and, crucially, to prolong the currency of the /ɡ/-less variant. The hypothetical proposal is, then, that today recognize can be pronounced without /g/ not because of some French interference but because of interference from reckon. Recognize then, when /'rekəgnaɪz/, is a case of non-transparent derivation from a bound stem recogn-, and when /'rekənaɪz/, it is a different case of derivation from the free stem reckon. This latter case should also be considered non-transparent, since reckon only lends its phonological form (and possibly some of its semantics too).