Distinctive features and sound systems

Sub-segmental phonological units

If we say that the voiced and voiceless lateral are two of the distinctive sequential atoms of Welsh, then surely we will also want to say that voiced and voiceless are two distinctive components or distinctive features which distinguish (and partly identify) these two phonemes. Treating a Welsh /l/ as a sequential atom should not, in other words, debar us from analysing the phoneme further into simultaneously occuring sub-atomic particles, one of which is the feature voiced. An English /l/ phoneme, as a distinctive unit, is of course not characterised as either voiced or voiceless, since the contrastive possibilities which such a characterisation implies are not open to it.

Now, obviously, the two laterals of Welsh are not the only Welsh phonemes, nor is the English lateral the only English consonant and it would seem to follow from this - in so far as each phoneme of a language must be distinguishable from each other phoneme in the language - that the Welsh and English lateral phonemes must be distinguished, as we cast our net wider within each of the languages, by rather more distinctive features than we have so far noticed.

Establishing the distinctive features

How, then, do we flesh out the distinctive feature characterisation of a given phoneme? Well, in just the same way as we established voiced and voiceless as distinctive features of the two Welsh l's. We pair the phoneme we are interested in, turn by turn, with all of the other phonemes within the same language and ask what feature or features guarantee the distinction between the members of each pair. Naturally, the really persuasive cases are those where only one feature keeps the phonemes apart. English /b/, for example, is kept distinct fom English /p/ by being consistently voiced as opposed to voiceless, but it is also distinguished from /d/ by being labial (not lingual ), from /m/ by being oral (not nasal), from /v/ by being plosive (not fricative). In English, therefore, the /b/ phoneme can be decomposed by this technique into a collection of distinctive features which includes at least voiced, labial, oral, and plosive. If you prefer, English /b/, distinctively speaking,is a voiced labial oral plosive.

As it happens, when we try to do the same thing with the English /l/ phoneme, there turns out to be rather less to say. Although the /l/ may indeed exhibit a number of consistent characteristics when we oppose it to other English phonemes, /l/ is guaranteed distinct from every other phoneme in the langauge as soon as we have identified it as lateral. Or to put it another way, if we know that a phoneme of English is a lateral then we can predict all of its other features. In that sense, they are redundant. If adopt Occam's razor, then, and use it to eliminate unnecesary entities, the distinctive feature characterisation of English /l/ should be no more than lateral.

In fact, postulating underdeveloped distinctive feature characterisations like that of English /l/, is as much about explanation as it is about economy. A phoneme which is sparsely identified ought to have room for phonetic manoeuvre denied to its more tightly constrained collegues. So it does and sometimes that freedom is strikingly flaunted. English /r/, for example is well enough distinguished from other English consonants when we have said that it is a non-nasal, non-lateral sonorant. It is this lack of tight determination which allows a wide range of phonetically different sound types to function as /r/ in different varieties of the language - RP, Scots, Estuary, Northumbrian, for instance. (How many different English r's can you identify? )

Features and the structure of sound systems

Thus far, we have treated phonemes as the primary elements of phonology and features as characteristics (or components) of phonemes. Suppose now that we turn the tables and assign primacy to the features, viewing them as forming the dimensions of some multi-dimensional phonological space within which the phonemes of a language are then able emerge.

Turkish vowels, for example, are, in this light, points in a phonological space which is defined by the following three dimensions or parameters:

If substitute the more everyday breadth for rounded -unrounded, then along with height and depth we clearly have here the set of dimensions we normally use to identify a box (or any regular three dimensional object). For that reason, the inter-relationships between the vowels of Turkish can be directly illustrated by picturing them as occupying the intersections at each of the eight corners of a cube:

Looked at in this way the Turkish vowels are seen to hang together in an extremely economical structure. With only three binary feature distinctions, eight phonemes (the maximum number) are kept clearly apart. Turkish was chosen to illustrate the idea because the vowel distinctions are particularly neat and tidy, but Turkish is not really very exceptional in this respect. If we tabulate the consonants of French within the space formed by the set of phonetic dimensions this language exploits, the symmetry of the pattern, though not as perfect, is none the less perfectly obvious.

The fact is that no matter what language you choose, you will find it economical in its exploitation of phonological space. To put it another way, the phonemes of a language are not members of some ad hoc inventory or some random collection of sounds, each one is an element with a well-defined place in a sound system.

Sound system universals

Sound systems are language specific, of course. The sound system of French is not like the sound system of Turkish. But, there are none the less interesting constraints on the form such systems seem able to take and on the ways in which they may diverge.

Some of these constraints are fairly certainly absolute, like: All languages distinguish consonants and vowels. Some are statistical, identifying the normal, the default or (as linguists often say) the unmarked case. We might claim for example that Vowels are normally voiced (though voiceless vowels are found) or that Plosives are typically voiceless (even though voiced plosives are hardly rare).

Many of the generalisations which can be made are, however, most naturally treated as implicational in type. If a language distinguishes fricatives by voice then it will distinguish plosives by voice. If a language has nasalised vowels then it will have oral vowels. Implicational relationships can be interpreted as conditions on the growth of phonological systems from simple to more complex structures. Growth occurs most obviously in the language development of an individual, but we also find growth (in size) as we survey any range of systems moving from smaller to larger. Implicational constraints of the type we have mentioned determine how such growth may be achieved as progressive feature differentiation develops the system. We would anticipate that growth in language development would mirror cosely the growth found in the divergence between languages.

Linguists who study such general patterns of sound systems would say that they work on phonological universals and typology.

Non-distinctive is not non-informative

This page has concentrated on the distinctive feature characterisation of phonemes. If we turn our attention to allophones, it is clear that these segments would need to be represented by a more complete feature specification to to take into account the systematic differences found in the different positions in which they occur. The status of these extra allophonic features is obviously very different that of the distinctive features. Because they are not distinctive in the strict sense, it is normal (and you might not think it uncontroversial) to call them non-distinctive features.

But beware of equating non-distinctive with non-informative. Features that are non-distinctive may not be used to distinguish words, but they are still available for exploitation in other ways. The way an English speaker hears Welsh demonstrates clearly that non-distinctive features are directly used in the decoding of utterances.

Another example will perhaps help to emphasize the point. We would no doubt claim that aspiration is not a distinctive feature of English voicless plosives, since aspirated plosives never appear within words in the same positions as unaspirated plosives. But despite that, aspiration can provide crucial information for the interpretation of sentences. It is the difference of aspiration of the t's in the pair of sentences Jack stumbled and Jack's tumbled which identifies the different position of the word boundary in each case. It is the unaspirated character of the /t/ in stumbled which signals that it is preceded in the same word by an /s/ and the aspirated character of the /t/ in tumbled which sets it firmly at the beginning of the syllable (and the word). In other words it is the non-distinctive(!) aspiration difference which in the end distinguishes the utterances.

Putting distinctive features to use

The idea that a segment (phoneme or allophone) may be represented as a collection - a bundle - of features, which reflect the place of the segment (relative to others) within the system as a whole, has proved very fruitful in linguistic research. As an example of the way in which we can take immediate advantage of the notion, as Logo programmers, take a look at a simple 'feature bundle microworld' which handles the descriptive duty of identifying the segment types available to speakers of a language by generating feature bundles which meet some appropriate set of feature co-occurrence restrictions. Links are provided from there to other Logo systems which follow up the distinction between phonemic and allophonic representations. If you are interested in exploring phonological system univerals in Logo, then try this page.

Ron Brasington
Department of Linguistic Science
The University of Reading

E-mail: ron.brasington@rdg.ac.uk