Distinctive Features - Speech Resource Pages - Macquarie University

Regardless of the many differences and controversies, the various kinds of
feature systems share the following characteristics.
a) Features establish natural classes
Using distinctive features, phonemes are broken down into smaller components.
For example, a nasal phoneme /m/ might be represented as a feature matrix
[+ sonorant, -continuant, +voice, +nasal, +labial] (nb. you will often see the
features of the matrices arranged in columns in phonology books -- this is exactly
equivalent to the 'horizontal' representation used here). By representing /m/ in
this way, we are saying both something about its phonetic characteristics (it's a
sonorant because, like vowels, its acoustic waveform has low frequency periodic
energy; it's a non-continuant because the airflow is totally interrupted in the oral
cavity etc.), but also importantly, the aim is to choose distinctive features that
establish natural classes of phonemes. For example, since all the other nasal
consonants and nasalised vowels (if a language has them) have feature matrices
that are defined as [+nasal], we can refer to all these segments in a single
simple phonological by making the rule apply to [+nasal] segments. Similarly, if
we want our rules to refer to all the approximants and high vowels, we might
define this natural class by [+sonorant, +high].
The advantage of this approach is readily apparent in writing phonological rules.
For example, we might want a rule which makes approximants voiceless when
they follow aspirated stops in English. If we could not define phonemes in terms
of distinctive features, we would have to have separate rules, such as [l]
becomes voiceless after /k/ ('claim'), /r/ becomes voiceless after /k/ ('cry'), /w/
becomes voiceless after /k/ ('quite'), /j/ becomes voiceless after /k/ ('cute'), and
then the same again for all the approximants that can follow /p/ and /t/. If we
define phonemes as bundles of features, we can state the rule more succinctly as
e.g. [+sonorant, -syllabic, +continuant] sounds (i.e. all approximants) become
[+spread glottis] (aspirated) after sounds which are [+spread glottis] (aspirated).
If the features are well chosen, it should be possible to refer to natural classes of
phonemes with a small number of features. For example, [p t k] form a natural
class of voiceless stops in most languages: we can often refer to these and no
others with just two features, [-continuant, -voiced]. On the other hand, [m] and
[d] are a much less natural class (ie. few sound changes and few, if any,
phonological rules, apply to them both and appropriately it is impossible in most
feature systems to refer to these sounds and no others in a single feature
matrix).
b) Economy
In phonology, and particularly in Generative Phonology, we are often concerned
to eliminate redundancy from the sound pattern of a language or to explain it by
rule. Distinctive features allow the possibility of writing rules using a considerably
smaller number of units than the phonemes of a language. Consider for example,
a hypothetical language that has 12 consonant and 3 vowel phonemes:
p t k
b d ɡ
m n ŋ
f s ç
i u
ɑ