Free variation

We dealt with an exception to the criterion of predictability of occurrence: two sounds which are in complementary distribution are normally assigned to a single phoneme, but where this would conflict with phonetic similarity (and with native speakers’ intuitions), it is appropriate to set up two distinct phonemes and seek an alternative explanation for the complementarity, in terms of defective distributions. In this section, we turn to an exception to the other main criterion for allophony, invariance of meaning.

When one sound is substituted for another and no meaning difference arises, we are dealing with two allophones of the same phoneme. An English speaker who produces a darkin initial position may be regarded as having an unfamiliar accent, or some sort of minor speech impediment, but there is little danger that light pronounced with initialis going to be mistaken for another word entirely.

However, sometimes there is more than one possible pronunciation in the same word or context; this is known as free variation, and raises two possible theoretical problems. First, we require complementary distribution to assign two sounds to a single phoneme; and yet a speaker of Scottish English, for example, may sometimes produce a tapped allophone of /r/ in very, and on other occasions, an approximant. There are no possible minimal pairs for tapped [ɾ] versus approximant [ɹ], and an allophonic rule can indeed be written, such that the tap appears intervocalically, as in very, and the approximant word-initially and word finally. Apparent exceptions are sociolinguistically motivated: perhaps the Scot is talking to an English English speaker, who will typically not use the tap, and is subconsciously accommodating her speech towards that of her interlocutor; perhaps she is trying to sound less like a Scot; perhaps she is in a very formal situation, where more standard pronunciations are favored. Clearly, such stylistic variation is not free in sociolinguistic terms, though it is known as free variation phonologically because there is no watertight phonological or phonetic context determining the appearance of one allophone rather than the other. The variable appearance of a glottal stop or [t] medially in butter, for instance, would fall into the same category, and the frequency of occurrence of the two variants would be subject to explanation in the same sociolinguistic terms.

The second type of free variation is the converse of the first, and potentially more problematic. Here, instead of finding two allophones of a single phoneme in the same context, violating complementary distribution, we see two sounds which on other criteria belong to different phonemes, failing to make the meaning difference we expect. Sometimes the difference can be explained in geographical terms: for instance, Southern British English speakers say tomahto, and North American speakers typically say tomayto, producing the same lexical item with consistently different vowels. Those two vowels, [ɑ] andrespectively, nonetheless contrast for speakers of both accents, although as we shall see in more detail in the next three chapters, they appear in different sets of words: a Southern British English speaker will have relevant minimal pairs in psalm and same, or grass and grace, while a General American speaker will contrast lot with late, or odd with aid. The two different pronunciations of tomato are therefore simply characteristic of speakers from different areas.

In other cases, the same speaker uses different phonemes in the same word on different occasions of utterance. Some speakers consistently pronounce economic with the [ε] of elephant, and others with the [i] of eat; but many more produce sometimes one, and sometimes the other. And yet there are plenty of minimal pairs to establish a contrast between /ε/ in pet, hell or bed, and /i/ in peat, heal or bead, outside that single problematic lexical item. The same is true for either and neither, which some speakers produce with [i], others with theof high, and still others with variation between the two. Again, there is no question that /i/ andconstitute different phonemes, with minimal pairs including he and high, heed and hide, or steal and stile. This is theoretically problematic: two sounds which on all other criteria belong to different phonemes are nonetheless found in the same context without making a meaning difference, directly contravening invariance of meaning. However, such examples tend to be few and far between, and involve only single lexical items; and again, the explanation is typically sociolinguistic. These pronunciations often develop in different geographical areas, then one spreads into the territory of the other. One variant may become stigmatized, and the other fashionable; but this stylistic variation can disappear over time, leaving two rather neutral alternatives. In such cases, the resulting variation can be truly free; but as long as the phonemes involved can be identified on the basis of minimal pairs elsewhere, these can simply be regarded as one-off exceptions. They are parallel to cases where a speaker stores two words, from the same historical source but each now appropriate in a different dialect, like the Scot who uses kirk with fellow Scots, but otherwise church; or indeed, to the use of historically unrelated synonyms like sofa and settee.