Andrew Radford - Linguistics An Introduction [Second Edition]

The way we will proceed is to assume (adopt the hypothesis) that the underlying representations for words like pit and spit do not specify whether the plosive is aspirated or not. After all, we don’t need this information in order to distinguish the two types of word, since aspiration is not a distinctive feature in English. Put differently, aspiration is a completely redundant feature because its distribution can always be predicted, unlike voicing, which serves to distinguish words like pit and bit. The way we indicate that a feature is redundant is to give it the specification ‘0’: [0aspirated]. We often say that such a sound is underspecified for the feature (for the use of a similar notion of underspecification in connection with children’s syntax, see section 24, p. 361). However, we can’t pronounce an underspecified sound (because we won’t know whether to aspirate the sound

or not), so ultimately we will need a rule which will specify various occurrences of

/p t k/ as [+aspirated] or [−aspirated]. The idea that some features are specified in underlying representations while other features are underspecified is very important because this is the main way of formalising the idea that some feature

specifications are contrastive in the language.

The aspiration rule is stated informally (i.e. in ordinary prose) in (65):

(65) a.

In /p t k/, [0aspirated] is given the specification [−aspirated] after s-.

b.

In /p t k/, [0aspirated] is given the specification [+aspirated] in other positions.

‘Specification’ is a process which we can symbolise using an arrow → (as we did in the case of vowel reduction). The notion ‘in a given position’ is symbolised by a slash which represents the environment or context in which the process occurs.

Incorporating these two pieces of notation into (65) gives us (66):

88

sounds

(66) a.

In /p t k/, [0aspirated] → [−aspirated] /s___

b.

In /p t k/, [0aspirated] → [+aspirated] / other positions.

The part of the rule in (66a) says that the phonemes /p t k/ are realised as the unaspirated allophones immediately after /s/, and (66b) says that they are realised as the aspirated allophones elsewhere. The line ___ in (66a) is called the focus bar. If the plosives had been aspirated whenever they preceded s (in the clusters -ps, -ts,

-ks), then the focus bar would have come to the left of the s in the statement of the appropriate rule. Recalling that we can use the IPA diacritic ‘=’ to indicate that a sound is unaspirated, we can say that the two rules in (66) are interpreted as in (67): (67)

The phonemes /p t k/ are realised (pronounced) as

a.

the allophones [p= t= k=] after s

b.

the allophones [ph th kh] elsewhere

Now, we can improve on the formulation in (66) in an important way by making use of distinctive features. Notice that the aspiration affects a specific group of sounds, the voiceless plosives. It isn’t an accident that aspiration affects these sounds and not others. For instance, the English aspiration process is a natural process, of a kind we might expect to see in other languages. But we can imagine dozens of other entirely unnatural processes affecting different hypothetical groupings of consonants, such as /p l n/ or /v g s/. However, it is only well-defined groups such as ‘voiceless plosives’ that undergo phonological processes. Such well-defined groups are called natural classes, and one of the most important functions of distinctive features is that they present us with a means of distinguishing natural from unnatural classes.

The set /p t k/ is exactly that set of sounds which simultaneously bear the

specifications [−voiced, −continuant]. All the other [−continuant] sounds (i.e.

stops such as /b/ or /n/) are voiced and all the other voiceless sounds are either continuants (the voiceless fricatives) or affricates (and hence [−/+continuant]).

On the other hand, a non-natural class such as /p l n/ can’t be represented in such simple terms. Thus, /p l n/ are all consonants, hence, [+consonantal] (see appendix

2 for this feature), but the [+consonantal] class includes all the other consonants too. The feature [−voiced] doesn’t apply to the whole set because /l n/ are voiced, but neither does [+voiced] because /p/ is [−voiced]. If you check against the feature matrix in appendix 3, you will see that there are no other features which members of this class have in common. This means that a characterisation of

this set in terms of features will be very cumbersome and will have to take the form of (68):

(68)

Feature characterisation of /p l n/:

[−voiced, −continuant, Labial]

(/p/)

OR

[+lateral]

(/l/)

OR

[+nasal, Coronal]

(/n/)

Phonemes, syllables and phonological processes

89

This crucially involves the use of the word ‘or’, which means that we have to resort to effectively listing the separate phonemes of the set. The set /p l n/ is thus like a set {milk, elephant, violin}: apart from the fact that the members of this latter are all physical objects, they have nothing in common. However, the set /p t k/ is more like the set {violin, viola, cello}, which is a natural grouping characterisable as ‘set of instruments used in forming a string quartet’.

It might be objected that we’ve weighted the scales by selecting an obviously unnatural grouping like /p l n/. But the same will be true of, say, /p t g/, which is at least a set of plosives, with only one member different from our natural class. This, too, however, can’t be described using features without resort to ‘or’, but this time it’s simply because /g/ is [+voiced], while the other two sounds are [−voiced].

Thus, a small change (in this case of one feature specification for one sound) can make all the difference between a natural class and a non-natural class. In a language like English, we wouldn’t expect /p t g/ to be involved in a phonological process to the exclusion of, say, /b d k/. Neither of these is itself a natural class, but /p b t d k g/ is, being exactly characterised as [−continuant, −nasal].

To return to aspiration, using the distinctive feature notation, we can rewrite (66)

as (69), where we have abbreviated the names of the features in standard ways: (69) a.

[−voiced, −cont, 0asp] → [−voiced, −cont, −asp] /s___

b.

[−voiced, −cont, 0asp] → [−voiced, −cont, +asp] /other positions

In practice, these rules can be further simplified by virtue of a notational convention which says that we don’t need to mention feature specifications on the

right-hand side of the arrow if they don’t undergo a change via application of the rule. This means that we don’t need to mention [–voiced, –cont]. Thus, we have (70):

(70) a.

[−voiced, −cont, 0asp] → [−asp] /s___

b.

[−voiced, −cont, 0asp] → [+asp] /other positions.

Finally, we now employ a further notational convention which allows us to

collapse the left-hand sides of the two subparts of (70). There are only two possible values for the feature [aspirated], so there are two subrules telling us how a voiceless plosive is pronounced, as shown in (71): (

)

(71) a.

[−voiced, −cont, 0asp] →

[−asp] /s___

b.

[+asp]

These two subrules are interpreted as follows: when we encounter a voiceless plosive which has no specification for [aspiration], we first look to see if it is preceded by /s/. If it is, then it is marked [−asp]. Under any other circumstances, it is marked [+asp]. This means that we must apply subrule (71a) before subrule (71b), because if (71b) applied first, it would incorrectly aspirate the voiceless plosive in a word like spit. However, there is a very important principle in linguistics which means that we don’t have to stipulate that (71b) follows (71a).