When studying phonology, it is fundamental to understand phonemes in terms of the features that they bear and share. Doing so allows us to group phonemes together into natural classes, which gives insight into how classes of sounds pattern together when undergoing various phonological processes. Natural classes are groups of phonemes that share one or a set of phonological features.
Natural classes are described by the minimum number of binary features [±] that all phonemes in the class bear, to the exclusion of all other sounds.
For example, the phonemes [p, t, k,] can be grouped together as a natural class by showing the binary distribution of the features in Table 1.
This distribution of features will distinguish these 3 sounds from all other phonemes in Standard English. Thus we can say that [p, t, k] are [-syll, -vce, -cont,-del rel]. All other phonemes in Standard English are excluded from this natural class. However, if one feature is altered, the natural class is changed.
In Table 2, when altering voicing from [-vce] to [+vce], the members of this new natural class changes such that we now include [b, d, g].
Table 3 shows that when one more feature is altered, yet a different natural class is yielded. Now that the feature [continuant] has been altered from [-] to [+], we must add all the sounds in English that fit this description. 2
In Table 3 we see that when [+cont] is specified rather than [-cont], the natural class is altered. Keep in mind that this is language-specific.
Table 4 shows the basic vowel inventory in Standard Spanish based on features.
Here is an exhaustible list of features used in building natural classes.
Often the case is that, the fewer features used to describe a natural class, the larger the class. For example, in most languages, one feature such as [-syll], will distinguish all consonants from all vowels (1a). Specifying one feature yields a large natural class. However, by adding features, we reduce the possible candidates for a natural class. For instance, if we add [+voice], then only voiced consonants are permitted in this class (1b). Furthermore, by specifying [+continuant] the natural class is reduced to only voiced fricatives (1c).
(1) Modern Persian
- [-syll] = [p, b, t, d, k, g, ʔ, f, v, s, z, ʃ, ʒ, tʃ, dʒ, x, ɣ, ɢ, h, ɹ, l, n, m, ŋ, j]
- [-syll, +voice] = [b, d, g, v, z, ʒ, dʒ, ɣ, ɢ]
- [-syll, +voice +continuant] = [v, z, ʒ, ɣ]
The more features required in building a natural class, the more complex the sounds. /t/ can be described with the features [-son, +ant, +cor, -cont]. Adding the features [+cont] and [+del rel] creates a more complex sound, [θ].
Linguistic theory therefore provides a simplicity metric, which will automatically assign simplicity coefficients to alternate solutions so that the correct solution is chosen. In this way the theory reaches the level of explanatory adequacy, that is, it motivates the choice of the best grammar from all the descriptively adequate grammars. (Hyman, 1974)
In generative phonology3, a minimum number of elements (phonemes, rules, features, etc.) is always preferred in analysis. Thus, in creating natural classes, it is crucial to specify only the features that correctly describe the set of phonemes without giving any more information than is required. For more on this topic, see our tutorial on Redundancy.
While each of the features in Table 5 provides a way of classifying phonemes, it is not necessary to include all features in a description rather only those that are relevant in distinguishing one set of sounds to the exclusion of all others. Some features overlap and some can be predicted from others; predictable features are referred to as redundant. In fact, binary features are mutually exclusive; a sound cannot be both [+high] and [+low] (though it can be [-high] and [-low]).
Just as languages differ in terms of types of phonemes, they also differ in which features are relevant. Languages have a default value for phonemes that need not be specified if they are otherwise predictable (by virtue of their membership in another feature class). These basic features are referred to as underspecified since they are understood to always be present. For example, in English [-back] vowels are also [-round] such that if a vowel is a member of the group [-back], it is automatically a member of the group [-round]. Thus [+round] is underspecified for [-back] vowels. There is no reason to specify roundness for [-back] vowels since only one option exists. Additionally, we need not specify that nasals are [+voice] since this is the default feature of [+nas] sounds. When we indicate that a sound is [+nas], we assume it to also be [+vce].
We can also use universal implications to predict the existence of one class of features based on the presence of another. For example, generally, if a language has a more complex version of a phoneme, it can be assumed that the less complicated version exists. The complexity of a phoneme is determined by several factors such as those that are most common cross-linguistically (across most languages) and those that children initially acquire. These are labeled as unmarked, or more natural, while the less common, later-acquired features are marked.
Examples include the following:
If /t/ then /d/ since a voiced consonant is considered more complex, or marked, than one that is voiceless. If /n/, then /m/ since a bilabial nasal consonant is considered more complex, or marked, than one that is alveolar. If /ɪ/, then /i/ since a lax vowel is considered more complex, or marked, than one that is tense.
Using distinctive features can be a good way to explore the phonemes of a language and gain insight into natural classes, both in specific languages and cross-linguistically.
- See The Basics About Distinctive Features for more information.
- Although the voiced fricatives in Table 3 are [+del rel], we only use this feature to distinguish stops from affricates, thus are not considering it relevant in this discussion.
- Generative Phonology is a component of Generative Grammar that accounts for the variation between the sounds in the mind of a speaker (underlying representation, or most simple form) and the phonetic representation (surface form), which represents the actual physical properties of the articulation of speech sounds, and the rules that account for such an alternation.
Hyman, L. Phonological Theory and Analysis. New York, New York: Holt, Rinehart and Winston, Inc.,1975.