Making statements more precise

The next question is how we should express these generalizations. Having established that certain sounds are allophones of the same phoneme, and that they are in complementary distribution, we might write a statement like (2) to say what happens to the phoneme or phonemes in question, and where.

These statements express the main generalization in each case. However, making a statement in normal English can be unclear and unwieldy, so phonologists typically use a more formal notation which helps us to work out exactly what is being said; it is easier that way to identify what a counterexample would be, and to see what predictions are being made. The English statement also does not tell us why /p/, /t/ and /k/ are affected, rather than just one or two of them; or why these three sounds should behave similarly, rather than /p/, /s/ and /r/, for instance. Similarly, we cannot see what /k/ and/g/ have in common, or indeed what the resulting allophones have in common, simply by looking at the phoneme symbols.

Introducing the articulatory descriptions , Immediately makes our statements more adequate and more precise, as we can now express what particular sets of sounds have in common (3).

We can take this one step further by regarding each of the articulatory descriptions as a binary feature: that is, a sound is either voiceless or voiced, and these are opposites; similarly, a sound is either nasal or not nasal. Instead of voiced and voiceless, or oral and nasal, we can then write [+voice] and [– voice], and [– nasal] and [+nasal]. This may seem like introducing needless complexity; but once you are used to the notation, it is much easier to compare these rather formal statements, and to see what the important aspects are.

These distinctive features allow each segment to be regarded as a simultaneously articulated set, or matrix, of binary features, as shown in (4).

These features, however, are not entirely satisfactory. They do describe phonetic characteristics of sounds; but we are trying to provide a phonological description, not a phonetic one, and one interesting phonological fact is that features and phonemes fall into classes. For instance, the matrices in (4) have to include values for all three of the features [stop], [fricative] and [approximant], despite the fact that any sound can be only one of these. Together, they provide a classification for manner of articulation; but (4) lists them all as if they were as independent as [nasal], [voice] and [alveolar]. Similarly, in (4) values are given for [labial] and [alveolar], and we would have to add [labio-dental], [dental], [postalveolar], [palatal], [velar] and [glottal] for English alone: but again, it is simply not possible for a single consonant to be both labiodental or velar, for instance, or both alveolar and labial. We are missing the generalization that together, this group of features makes up the dimension of place of articulation.

One possible way of overcoming this lack of economy in the feature system is to group sets of features together, and write redundancy rules to show which values can be predicted. Redundancy rules take the shape shown in (5).

The first rule says ‘if a segment is a stop, it cannot also be either a fricative or an approximant’. All these redundancy rules are universal – that is, they hold for all human languages, and are in a sense statements of logical possibilities. Particular languages may also rule out combinations of features which are theoretically possible, and which may occur routinely in many other languages. Two language-specific redundancy rules for English are given in (6): the first tells us that English has no palatal nasal (although Italian and French do), and the second, that English has only lateral approximants (though Welsh, for instance, has also a lateral fricative). These redundancy rules cannot be written the other way around: it would not be accurate to say that non-palatals are all nasal in English, or that all approximants are lateral.

While we should expect to have to state redundancy rules of the sort in (6), since these express quirks of particular languages, it seems unfortunate that our feature system is not structured so as to factor out the universal redundancies in (5). However, to produce a better phonological feature system, we first need to spell out what we want such a system to achieve.