In this chapter we’ll start our investigation of the structure of language by looking at how the sounds of speech – the noises you made reading the previous chapter – are organised and exploited as part of the business of building linguistic units. The chief concept we’ll discuss here is the phoneme, our first truly abstract linguistic unit. Once we have seen what phonemes are and how they work, we can look at two further abstract entities: natural classes, which group phonemes together, and distinctive features, which make them up. Finally, we’ll take a very brief look at phonological rules and how they work.

In the previous chapter, we saw in some detail how complex even the most banal bits of everyday language are. Our little five-second, sixty-eight-sound sentence involves a great deal of carefully co-ordinated writhing and wriggling of the speech organs – impressive enough in itself – and manages to mean something. We’ll get to how it actually means something later (see Chapter 5). In this chapter, we’re going to look at how speech sounds are structured: this is the branch of linguistics known as phonology.

Be sure to get your terminology right: phonetics is the study of the speech sounds themselves; phonology is the study of how languages organise speech sounds into structured systems. To the extent that everyone has the same organs of speech (and the same perceptual apparatus, something we did not really go into in the previous chapter), phonetics is the same everywhere. But different languages may well deploy the same sounds differently, so English phonology and French phonology may be, and in fact to a fair extent are, somewhat different.

The main structural unit in a given language’s phonological system is the phoneme. Phonemes are the units of sound that make a difference in a language. We can isolate a language’s phonemes by setting up minimal pairs of words which contrast only in a single phoneme but which are clearly distinct words in terms of their meanings. So, for example, pie and buy are a minimal pair in English, contrasting only in the initial [p] in pie ([paɪ]) vs initial [b] in buy ([baɪ]). So this minimal pair shows us that /p/ and /b/ are phonemes of English.

The minimal contrast between tie and die shows /t/ and /d/ are also phonemes of English. Similarly for shy and sigh, the minimal contrast indicates that /ʃ/ and /s/ are also phonemes. And so on.

It’s very important to see that phonemes are not sounds. They are abstract linguistic entities that organise sounds. Take the English /p/ phoneme, for example. Pronounce pie and spy with your hand in front of your mouth. In the release phase of the <p> in pie you can feel the explosive puff of air on the back of your hand. This is because the diphthong [aɪ] doesn’t start immediately as the lips open, and so there’s a split second when all that’s happening is air coming out from the vocal tract. This is a very short [h]-sound, known as aspiration. In careful phonetic transcription, this pronunciation of /p/ is written [p^h]. In spy on the other hand, there’s no aspiration, as you can tell from the fact that there’s scarcely a puff of air discernible if you put your hand in front of your mouth. This is unaspirated [p]. You can see the difference between the two in the spectrogram in Figure 2.1.

Figure 2.1Spectograms of Normal British English Pronunciation of pie and spy

Spectrograms of a a pie and a spy; the lack of aspiration following the silent phase of the articulation of [p] in a spy can be seen by comparing the two images. The high-frequency noise of [s] is also clearly visible before the silent phase of [p] in a spy.

Spectrograms of a a pie and a spy; the lack of aspiration following the silent phase of the articulation of [p] in a spy can be seen by comparing the two images. The high-frequency noise of [s] is also clearly visible before the silent phase of [p] in a spy.

(Source: Phonetics Laboratory, University of Cambridge)

English has another kind of p-sound, found at the ends of words in normal, colloquial speech. Pronounce lap or tap as naturally as you can. You can either release the /p/, and you’ll get aspiration (try the back-of-the-hand test again). Or you can just leave your lips closed, giving what’s called unreleased /p/ (phonetically [p ̚ ]). So there are three ‘p-sounds’ in English, which are not contrastive, i.e. they don’t form minimal pairs (lap means the same thing whether you release, or aspirate, the /p/ or not). These different ‘p-sounds’ (and these are sounds, not phonemes) are known as the allophones of the phoneme /p/. So we say that English /p/ has three allophones: [p^h], [p] and [p ̚]. Moreover, each allophone shows up in its own special context: [p^h] at the beginning of syllables where there’s no /s/ in front of it, [p] when there is an /s/ in front of it and [p ̚] at the ends of words. The context conditions the possibility of the occurrence of a given allophone, and so this is known as conditioned allophony. Since [p^h] and [p] each have their specific context (at the beginning of a syllable after /s/ or not), we say that they are in complementary distribution. If two sounds are in complementary distribution, they cannot minimally contrast, and so they cannot be separate phonemes.

Remember that different languages organise their phonemes and allophones differently. For example, French has a /p/ phoneme (as in poids ‘weight’, paix ‘peace’ and prix ‘price/prize’, etc.). But there is no aspirated allophone: at the beginnings of syllables, i.e. in words like the ones just given, French /p/ is just [p]. In some languages, such as Hindi, aspiration of stops is contrastive, and so /p/ and /p^h/ form minimal pairs, as can be seen from examples, such as p^hal, meaning ‘knife edge’ and pal, meaning ‘take care of’.

Another example of conditioned allophony, and consequent complementary distribution of allomorphs, in English, concerns the /l/ phoneme (you can tell this is a phoneme as it enters into minimal pairs, such as lay and ray, etc.). Pronounce leaf: here, as we saw in Chapter 1, you should feel your tongue blade against the alveolar ridge and the air escaping over the lowered sides of the tongue. Now pronounce feel. Here things are the same in the front part of the oral cavity, but you should be able to feel the back of your tongue bunching up and raising towards the velum, in a position somewhat similar to that in the pronunciation of [u:] or [ʊ] (if you are Irish, American or a Londoner, these differences may not hold, or may not hold in the same way). The two ‘l-sounds’ are clearly different from a phonetic point of view. The first, in leaf, is called palatalised /l/, IPA [ļ], as it involves raising the tongue towards the hard palate. The second, in feel, is known as velarised /l/, IPA [ł], since it involves raising the tongue towards the velum.

In Standard British English, [ļ] and [ł] are never contrastive: [ļ] appears at the beginning of a syllable and [ł] at the end. So this is a further example of conditioned allophony giving rise to the complementary distribution of allophones of a single phoneme. The English phoneme /l/ has the two allophones [ļ] and [ł]. Again, other languages do things differently. Russian has distinct /ļ/ and /ł/ phonemes, as in ugol ‘corner’ (velarised /ł/) and ugol’ ‘(char)coal’ (palatalised /ļ/). Nonetheless, the palatalised /ļ/ isn’t quite the same phonetically as the English [ļ] allophone of /l/.

We’ve seen examples of other languages having phonemes that correspond to English allophones. We can also find the opposite situation: English phonemes that are allophones of a single phoneme in other languages. In English, /d/ and /ð/ are phonemes, as the minimal pair udder (/ʌdə/) vs other (/ʌðə/) shows. But in Spanish, [d] and [ð] are conditioned allophones of /d/, with [ð] occurring in all positions except at the beginning of words, which is reserved for [d]. So, in dos (‘two’), the first consonant is phonetically [d], while in Madrid, both <d>’s are pronounced [ð].

As I said, phonemes are not sounds. As such, we could in fact choose to write /l/ and /p/ as /3/ and /4/, or /@/ and /#/, or /Mary/ and /Dick/. But since they are ways of organising sounds in a linguistic system, we represent them with letters, and, as I already mentioned, the IPA symbols are convenient because they eliminate the complexities of spelling.

The principle of contrast, enshrined in the minimal-pair technique, is central to defining and discovering the phonemes of a language. However, sometimes the system doesn’t seem to mind which of a pair of sounds is used. We saw this at the allophonic level with the realisation of /p/ in lap and tap. Here the /p/ can be unreleased, and this is probably the most natural in spontaneous, colloquial speech. However, you can be careful (and mind your p’s, if not your q’s!) and pronounce it with release, and, indeed, aspiration. This optionality is called free variation.

We also find free variation at the phonemic level. Since phonemes define distinct words, given the principle of contrast, free variation of phonemes is idiosyncratic to certain words. An example of this is the pronunciation of economics, whose first vowel for some people can be either /ɛ/ or /i:/. We know that /ɛ/and /i:/ are phonemes, given the minimal contrast between bed and bead, so here we have free variation. Another example, at least in my English, is the pronunciation of either and neither, where <ei> can be either /aɪ/ or /i:/. Again, /aɪ/ and /i:/ are both phonemes, as the contrast between pie and pea shows.

So that’s the phoneme, the principal unit of phonological structure and the main element making up the contrastive sounds of words, which allows us to distinguish words. Phonemes are a pretty important part of linguistic structure. Table 1.3 in Chapter 1 listed the consonant phonemes of English (except for [Ɂ], which is not a phoneme but an allophone of /t/). The vowel phonemes, in a conservative variety of Southern British English, are listed below in Table 2.1.

Phonemes can be grouped together in classes, or natural classes of sounds. Very often, important generalisations about a language’s phonological system can be made in terms of natural classes. In fact, we have been implicitly talking about natural classes all through this chapter and the previous one. Two very large natural classes are vowels and consonants. One generalisation about consonants is that they only appear at the edges of a syllable. We could also say that vowels only appear in the middle part of a syllable, but on the other hand we could say that at least high vowels can appear anywhere in a syllable: when /i:/ and /u:/ appear at syllable edges we call them the semi-vowels /j/ and /w/. Also the bizarre English /r/ is, phonetically, not a consonant because it doesn’t in any way obstruct the air stream. But phonologically, it is a consonant because it only appears at syllable edges. Since different languages organise their phonemes differently, we might expect to find a language in which a sound phonetically very similar to the English [r] really acts like a vowel; this is true of Mandarin Chinese, for example.

Another generalisation about these natural classes is that vowels are always voiced. Consonants, of course, can be voiced or voiceless, as we saw in the previous chapter. In describing the phonetics of vowels, I left this point implicit, but you can easily verify it with the finger-on-larynx test. Voiceless vowels are certainly phonetically possible: just pronounce any vowel and switch the larynx off; what you’ll hear is something like an [h], the sound of air passing through the open glottis. Moving the tongue and lips to form different vowel shapes can give this [h] a hint of different vowel qualities. But they’re hard to hear and, presumably for this reason, very rare in the world’s languages. The Amerindian language Cheyenne has quite a few voiceless vowels though (at least phonetically); you can hear them being pronounced at www.youtube.com/watch?v=JaWvsONEEno.

Looking at vowel classes, and leaving the complexities of diphthongs aside, we can discern some further natural classes. One is front vowels; another is back vowels. A further one could be rounded vs unrounded vowels. But look at the front vowels of English: /i:, ɪ, ɛ, æ/. They are all unrounded. Now look at the back vowels: /u:, ʊ, ɔ:, ɒ/, all rounded. So, in English (and in many other languages), the following generalisation holds: all and only back vowels are rounded. This is not true in all languages: in French, for example, there are three front rounded vowels, found in the words lune (‘moon’), peu (‘a bit/few’) and jeune (‘young’). The IPA symbols for these vowels are /y/, /ø/ and /œ/, respectively. There are also languages with back unrounded vowels, such as Turkish, in words like ılık (‘lukewarm’); note the dotless <ı> here, which is how this vowel is spelled in Turkish. The IPA symbol is /ɯ/.

Among the consonants, nasals are a natural class. They, too, are always voiced in English and most other languages (do the finger-on-larynx test again to see this). Voiceless nasals exist in some languages. In Welsh, for example, the <nh> in fy nhad (‘my father’) is a voiceless [n], written $\left[\begin{array}{c}\mathrm{n}\\ \mathrm{˳}\end{array}\right]$ in IPA (when there isn’t a separate symbol for a voiceless element, the IPA writes a little ˳ under the letter).

Other natural classes of consonants are stops, fricatives, velars, alveolars etc. You might notice that these classes cross-cut. So /t, d, n, s, z, l/ are all alveolars of one kind or another, while /b, p, t, d, k, g/ are all stops. This cross-cutting suggests that we might be able to break the phonemes down into smaller elements and then combine and recombine these smaller elements to form natural classes.

These smaller elements are known as distinctive features. These are features which are capable of contributing to phonological distinctions. Distinctive features define natural classes, in that natural classes are groups of phonemes which share a specification for one or more distinctive features. For example, the natural class of voiceless consonants of English (and many other languages) can be defined by the single feature [-voice] (since vowels are always voiced in English, this feature on its own singles out all and only the voiceless consonants).

If distinctive features are the atoms of phonology, then phonemes are the molecules. Distinctive features have a number of important properties. First, they are phonetically based, in that most distinctive features indicate an articulatory dimension which can vary independently of others. Sometimes this phonetic dimension has to do with the perception rather than the articulation of the sound. So, although, as we have seen, phonemes are not sounds, they can be defined as abstract entities in terms of characteristic phonetic properties of their allophones. Second, distinctive features are thought to be universal, i.e. to be the same for all languages. This is obviously linked to the previous point since we have already observed that the vocal tract is the same everywhere. Third, distinctive features are defined in binary terms, so they are normally preceded by a ‘+’ or a ‘-’ sign, indicating the presence or absence of whatever property it is. Representing distinctive features in binary terms in this way is handy and simple for the linguist, but it may represent a deeper kind of simplicity in the way the brain is organised (I’ll come back to this point at the end of the chapter).

So let’s look at a few distinctive features. The first one to look at is [±syllabic]. If a phoneme or an allophone is [+syllabic], it is able to form a syllable, either to be the middle part (or nucleus) of a syllable, or in fact to form a whole syllable on its own (in which case we could still regard it as a nucleus, just one lacking any edge-material). If a sound is [-syllabic], it cannot do this.

This feature creates our first two natural classes: vowels can be syllable nuclei (i.e. they are the middle part of syllables, and they can be syllables on their own). So /ɑ:/ (spelt are), /ɔ:/ (spelt or), for example, are syllables all on their own, but /b/ and /z/ are not. So vowels are [+syll] and consonants are [-syll]. Semi-vowels and /r/ are [-syll] since they are always at the edges of syllables.

Next, there is [±voice]. A sound is [+voice] if it is produced with the vocal cords vibrating and [-voice] otherwise. So all vowels are [+voice]. More succinctly, [+syll] implies [+voice]. Now you can begin to see how distinctive features allow us to express generalisations about sound systems in a neat way.

If a sound is [+nasal], it is pronounced with the velum lowered. Otherwise it is [-nasal]. So, we can say that [+nasal] implies [+voice]. Since in English there are no nasalised vowels, [+nasal] also implies [-syll]. This last implication does not hold true in French, though, since French has nasalised vowels.

Distinctive features apply equally to consonants and vowels. So, for example, [+back] means that the tongue moves to the back of the mouth. So, back vowels like /u:, ʊ, ɔ, ɒ/ are [+back]. But so are velar consonants like /k, g, ŋ/, as well as the semivowel /w/. The distinctive feature [+round] defines phonemes articulated with liprounding; [-round] applies to all others. So, for English, [+round] implies [+back], but not for French, as we saw.

The features [±low] and [±high] indicate the position of the tongue in the vertical dimension in the oral cavity. Again, these features apply to both consonants and vowels. So /i:, ɪ, u:, ʊ/ are [+high], and so are /k, g, ŋ, j, w/.

There are many more distinctive features but not enough space here to take you through all the definitions and examples. Tables 2.2, 2.3 and 2.4 give a full list of the distinctive features relevant for English, and the values the English phonemes have for each one. In order to understand these tables fully, you need to know about two more distinctive features: [±sonorant] and [±consonantal]. A [+sonorant] sound does not involve audible turbulence in the vocal tract while a [-sonorant] one does. So the [-sonorant] phonemes of English are the stops and the fricatives (i.e. the obstruents, as introduced in Chapter 1) but not the nasals. Consonantal segments are produced with an audible constriction in the vocal tract. The difference between [+consonantal] and [-sonorant] is that [+consonantal] includes, in addition to the obstruents, the nasals and liquids. Vowels and semi-vowels are [-consonantal].

Here you can see that while [+high] logically enough implies [-low] and [+low] implies [-high], [-high] and [-low] can go together. The mid vowels /ɛ, ɜ:, ʌ, ə/ are all [-high. -low]. You’ll also notice the [±tense] feature, which distinguishes tense /i:, u:/, for example, from lax [ɪ, ʊ]. We briefly looked at the phonetic notion of tenseness in the previous chapter.

Most of the distinctive features in Table 2.3 are self-explanatory. The feature [±continuant] distinguishes sounds with closure in the oral cavity from all others so nasals are [-continuant] as they have such closure (although of course the air goes out through the nose, as we saw in Chapter 1). The feature [±coronal] distinguishes alveolars from everything else while [±anterior] defines any consonant articulated at or in front of the alveolar ridge. So bilabials and labio-dentals are [-coronal, +anterior]. The feature [±spread] has to do with the position of the vocal folds and distinguishes /h/ from everything else. I’ll explain [±high] and [±back] as they apply to consonants below.

In Table 2.4 we see that [±cont(inuant)] distinguishes just the sounds that can be ‘held’, in the sense we illustrated in Chapter 1, from those which cannot be held. In other words, it distinguishes stops ([-cont]) from fricatives ([+cont]). Here [±anterior] distinguishes alveolars, such as /t, d, s, z/, from the post-alveolars /ʃ, ʒ/.

Distinctive features can be used to formulate phonological rules and to explicate how these rules apply to natural classes of sounds. Let’s look at a simple example of this in detail.

In English, the regular past-tense ending on verbs is written <(e)d> (the brackets around the ‘e’ mean it’s not always there in the spelling). So we have pairs like play (present), played (past); race (present), raced (past); want (present), wanted (past). Semantically the ending is always the same; it tells you the verb is referring to something that happened in the past (roughly, the past tense can also refer to unreal situations, as in If I had a hammer . . .). The spelling is also the same, apart from the <e> popping up or not here and there. But, now that you know some phonetics, think carefully about the pronunciation of these endings. The ending is pronounced, unsurprisingly, as /d/ in played. In IPA, we write play as /pleɪ/ and played as /pleɪd/. No surprises there. But in raced the ending is pronounced /t/: race is /reɪs/ and raced is /reɪst/. And, very clearly, in wanted the ending is pronounced /ɪd/: want is /wɒnt/ and wanted is /wɒntɪd/. So what’s going on? Something phonologically very neat, as we’ll see.

Let’s first look at some more verbs. Among the verbs that pattern like play in having a /d/ past-tense ending are raze, slave, bug, bang, ban, bomb, robe, breathe and bawl. Make up the past-tense forms, pronounce them and you’ll see.

Verbs that act like race, and take a /t/ in the pronunciation of the past-tense ending, include chafe, bank and rape. And verbs that act like want, which are less common, include word (as in she worded the letter carefully, with worded pronounced /wɜ:dɪd/).

Now, in distinctive-feature terms, what’s the difference between /t/ and /d/? The feature that distinguishes these two is of course [±voice]: /t/ is [-voice] and /d/ is [+voice] (see Table 2.4). Look again at the verbs which pronounce their past-tense endings as /d/. They all end in a [+voice] phoneme: a vowel (play), a nasal (bang), a sonorant (bawl), voiced stop (bug) or a voiced fricative (breathe). On the other hand, the verbs which take past-tense /t/ all end in a [-voice] consonant: a fricative (race) or a stop (bank).

So we could say that the past-tense ending is ‘really’ a /d/, but it turns into a /t/ when immediately preceded by a voiceless consonant. In other words, the final voiced phonemes of regular verbs like to be next to a voiced past-tense ending while the final voiceless ones like to be next to a voiceless past-tense ending. This general phenomenon of like going together with like is known in phonology as assimilation. The English past-tense endings we have seen here show voicing assimilation.

This rather strange business of one phoneme turning into another can be captured by phonological rules. So here’s Rule One:

As in a chemical formula or an algebraic equation, rules like this express a great deal of information in a very succinct and pared-down way; this is why at first sight they might seem rather imposing. However, they are fundamentally quite transparent; the key thing is to go through them very carefully and systematically step by step, or, more precisely, symbol by symbol. So, Rule One says that a phoneme specified as [+voice] turns into (this is the arrow) its [-voice] counterpart, i.e. all that changes in the specification of the phoneme is the value of the feature [voice], from ‘+’ to ‘-’. That’s the first part. The second part, after the oblique slash ‘/’, states the context in which [+voice] is to change to [-voice]. This is indicated by the ‘slash-and-dash’ notation (/ and ___), which states ‘in the context of’ (that’s the slash), and there’s the preceding context indicated by the first symbol after the slash; here that’s the single ‘#’. This single ‘#’ right after the slash indicates the beginning of the ending while the double ‘##’ after it indicates the end of the word. The dash (‘__’) in between indicates the position of the phoneme whose [voice] feature is to change from ‘+’ to ‘-’, in this case between ‘#’ and ‘##’. Together this contextual specification makes sure the rule only applies to endings, rather than applying to the middle of a simple word (we’ll say much more about endings when we look at morphology in the next chapter). In this way, Rule One states that a voiced phoneme turns into a voiceless one (at the same place of articulation and with all other features constant, as the rule refers only to [voice]) in an ending attached to a word ending in a voiceless consonant. In other words, the past-tense ending of a verb, such as race, chafe or bank, is pronounced /t/, although it is still a version of the past-tense marker /d/. Thus, /d/ changes into /t/ in exactly this context. Rule One captures this long-winded explanation with just a few concise abstract symbols (think again of chemical formulae or algebraic equations).

Now look at where the past-tense ending is /ɪd/. This is just where the verb ends in a /d/ (word) or a /t/ (want). So we could say that the past-tense /d/ turns into /ɪd/ when the verb ends in /t/ or /d/, i.e. when the verb ends in an alveolar stop. In distinctive-feature terms, alveolar stops are [-son, -cont, +ant(erior)], as Table 2.4 shows. An important aspect of phonological rules is that we always try to give the minimal feature specification that will clearly define the natural class we are interested in; in Table 2.4, you’ll see that [-son, -cont, +ant] are enough to single out /t/ and /d/.

If the past-tense ending is ‘really’ /d/, as we have suggested, then it too is an alveolar stop: [-son, -cont, -ant, +voice]. It looks as though a little /ɪ/ pops up between two alveolar stops, the final /t/ or /d/ of the verb, and the /d/ of the ending. This phenomenon of phonemes appearing where they are otherwise not seen is known as epenthesis. So here we have an epenthetic /ɪ/.

In terms of phonological rules, we can express this epenthesis operation as Rule Two:

Rule Two is a bit more complicated than Rule One, but we can again go through it step by step to see how it works. It says that nothing (or zero, written Ø) turns into (the arrow again) /ɪ/ in the context (indicated by the slash /) in between (indicated by the ___ in between the two lots of distinctive features) two alveolar stops (specified by the three distinctive features [-son, -cont, +ant]). As in Rule One, the ‘#’ and the ‘##’ after the slash ensure that the rule only applies to word-endings. (Of course, strictly speaking, /ɪ/ should be written in distinctive features, but I have spared you that detail). Saying zero turns into /ɪ/ is equivalent to saying that /ɪ/ is inserted; the slash and dash tell you exactly where /ɪ/ is inserted (in an ending before an alveolar stop). So Rule Two says that, where the verb ends in an alveolar stop, /ɪ/ appears in the ending. In other words, the past-tense ending /d/ ‘turns into’ /ɪd/ where the verb ends in /t/ or /d/. Since [voice] isn’t specified anywhere in the rule, the rule applies where the alveolar stops in question are either /t/ or /d/. Remember that anything which isn’t specified in the rule is left open; the rule states exactly what is to be changed.

So now we’ve got two phonological rules, very succinct and clever ways of expressing phonological generalisations using distinctive features to designate natural classes. But now we can go a step further and make our two rules interact with one another. To see this, let’s try putting Rule One and Rule Two in order. First, try what seems like the obvious order, Rule One before Rule Two. For example, we start from the verb want, i.e. /wɒnt/, and the ‘real’ form of the past-tense ending, /d/, and mechanically apply the two rules as they are stated above:

This is an example of a phonological derivation, taking us from an underlying form (the ‘real’ version of the ending, as we have been calling it up to now) to a phonemic representation of the actual pronunciation. The derivation proceeds by a set of ordered phonological rules. The order of the rules is very important and can actually help us to state the rules in the most economical form. To see how this works, look back at how we formulated Rules One and Two and apply them completely mechanically, paying close attention to how the distinctive-feature specifications isolate the classes of sounds that the rules apply to. Then you’ll see that Rule One will convert /d/ to /t/ since /wɒnt/ ends in a /t/ and Rule One converts a [+voice] phoneme to a [-voice] one in the context of an ending. Hence, it makes /d/, which is [+voice], [-voice] here, and the [-voice] version of /d/ is /t/. Now Rule Two comes in, and inserts epenthetic /ɪ/ between the two alveolar stops, /t/ at the end of /wɒnt/ and /t/ in the ending as modified by Rule One. So we get /wɒntɪt/ as the past tense of want, which is wrong. So this derivation, where Rule One is ordered before Rule Two, doesn’t work.

So let’s try switching the order of the rules. Let’s apply Rule Two before Rule One. Now look:

In this derivation, Rule One cannot be applied (or we can say that it applies vacuously; in other words it doesn’t do anything, as indicated here). This is because Rule One, as we have stated it, makes a [+voice] into a [-voice] immediately following a [+voice](at the beginning of an ending, as we saw). But, after Rule Two has been applied, /d/ immediately follows /ɪ/. Note that the ‘immediately’ specification here is crucial, and this is guaranteed by the way the rule is written. Since we haven’t added anything more to the context, each symbol in the context is taken to be immediately adjacent, or right next to, the ones on either side of it. As /ɪ/ is a vowel, it’s always voiced, and so Rule One has nothing to do here; it just doesn’t apply in the context created by Rule Two. So /d/ stays /d/ in the ending and we get /wɒntɪd/ as the past-tense of want, which is what we wanted. Rule Two, on the other hand, has no problem given this order of the rules: it inserts epenthetic /ɪ/ between the alveolar stop /t/ at the end of /wɒnt/ and the past-tense ending /d/. We get the right result if we apply Rule Two first because then Rule Two effectively prevents Rule One from being applied. Phonologists call this bleeding order: Rule Two ‘bleeds’ Rule One.

All of this footling about with the exact form of the past-tense ending might seem like a lot of work for not much of a result, but two important things emerge. First, formulating phonological rules in distinctive-feature terms gives us a very precise, succinct way of capturing generalisations about natural classes. Here the natural classes in question involve voiced consonants (expressed by [+voice]) and alveolar stops (expressed by [-son, -cont, +ant]). If we can successfully capture generalisations in this way then, as with chemical formulae, our phonological rules are telling us that the behaviour of linguistic sound systems is law-like, and so should be studied scientifically. Second, the rules interact in the context of the phonological derivation. We saw that Rule Two must be applied first, before Rule One gets a chance, or otherwise things go wrong. Actually there’s something even more intriguing here: you can easily see that Rule Two is more complicated than Rule One, in the simple and obvious sense that it involves more distinctive features (and so applies to a smaller natural class of sounds). So Rule Two is more specific than Rule One. Linguists have long thought that there may well be a very general organising principle in language which basically gives the specific precedence over the general. Rule Two getting to apply ‘first’ in the derivation – before Rule One – in determining the phonological shape of the English past-tense ending is a small example of this very general principle. If this is true, it is important because there is no obvious reason why it should be true; things could easily have been otherwise. Here we glimpse what might be a true law of language.

*

In this chapter, we have begun looking seriously at the abstract structure of language. Several abstract concepts have been introduced: phonemes, allophones, distinctive features, natural classes and phonological rules. Of these, the really central one is distinctive features, since phonemes, allophones and natural classes can all be defined using distinctive features, and phonological rules use distinctive features to describe phonological processes such as assimilation and epenthesis.

But what are distinctive features and phonological rules really? We could say that they are just a handy way of describing and summarising the nature of processes such as assimilation and so on. But there’s a more interesting and bolder answer: these formalisms and abstractions are an attempt to describe what the mind is doing when it structures the sound system of a language. We could maybe think of them as lines of code in the mental computer program for language; if so, then it’s no surprise that distinctive features are binary. This makes things very intriguing. Remember how impressive the intricate moves of the speech organs were in our five-second, sixty-eight-sound sentence in the last chapter? If this view of phonology is right, then, in addition to controlling the movements of the articulators (a nifty enough feat in itself), the brain is also in some way manipulating phonological rules and distinctive features all the time. So the feat of pronouncing a simple sentence becomes still more impressive! What makes phonology so interesting is that it is the part of linguistics where the brain of abstract structures meets the brawn of speech.

But we still haven’t said anything about how the noises, however they are made and whatever organises them, get to mean something. Now it’s time to start doing that; in doing so, we’ll meet a really major design feature of language.