Discovering phoneme

No-one has teach us, as children, how to discover the phonemes of our language. We do it unconsciously and at an early age know what they are. Before reading this book, or learning anything about phonology, you knew an l sound was part of the English sound system, a phoneme in English, because it contrast words such as leaf and reef. But you probably did not know that the l in leaf and the one in feel are two different sounds. There is only one /l/ phoneme in English, but more than one l phoneme. The /l/ that occurs before back vowels and at the end of words is produced not only as a lateral but with the back of the tongue raised toward the velum, and is therefore a velarised l. (Without more training in phonetics you may not hear the difference; try to sense the difference in your tongue position when you say leaf, lint, lay, let as opposed to lude, load, lot, deal, dill, dell, doll.)

The linguist from Mars, referred to in chapter 3, who is trying to write a grammar of English, would have to decide whether the two l sounds observed in English words represent seperate phonemes or are allophones of a single phoneme. How can this be done ? How would any phonologist determine what the phonological system of a language is ?

To do a phonemic analysis, the words to be analysed must be transcribed in great phonetic detail since you don’t know in advance which phonetic features are distinctive and which are not.

Given these words, do the voiceless/voiced alveolar stops [t] and [d] represent different phonemes or are they allophones of the same phoneme ?

Now consider the data from Greek, concentrating on the following sounds, three of which do not occur in English :

 

Summary

Part of one's knowledge of a language is knowledge of the phonology or sound system of that language the inventory of phones, the phonetic segments that occur in the language, and the ways in which they pattern. it is this patterning that determines the inventory of phonemes the segments that differentiate words.

Phonetic segment are enclosed in square brackets, [ ], and phonemes between slashes, / /. when phones occur in complementary distribution, they are allophones predictable phonetic variants of phonemes. For example, in English, aspirated voiceless stops such as the initial sounds in the words pill, till, and kill are complementary distribution (never occur in the same phonological environment) as the unaspirated voiceless stops following the s /s/ in spill, still, and skill; thus the aspirated p, t and k ([ph], [th], [kh]) and the unaspirated [p], [t], and [k] are allophones of the phonemes /p/, /t/, and /k/, respectively. On the other hand, phones which occur in the same environment and which differentiate words, such as the [b] and [m] in beat [bit] and meat [mit] represent two distinct phonemes, /b/ and /m/.

Some phonemes may be allophones of more than one phoneme. there is no one-to-one correspondence between the phonemes of a language and their allophones. In English, for example, stressed vowels become unstressed according to regular rules and ultimately reduce to schwa [ə], which is an allophone of each English vowel.

Phonological segments phonemes and phones are composed of phonetic features such as voiced, nasal, labial, and continuant, whose presence or absence is indicated by + or - signs. they distinguish one segment from another. when a phonetic feature causes a word contrast, as nasal does in boat and moat, it is a distinctive feature, whereas [+- aspiration] is not.

When two words (different forms and different meanings) are distinguished by a single phone occurring in the same position, they constitute a minimal pair. Some pairs, such as boat and moat, contrast by means of a single distinctive feature, in this case, [+- nasal], where /b/ is [- nasal]. Other minimal pairs may show sounds contrasting in more than one feature, for example, dip versus sip, where /d/, a voiced alveolar stop, is [+ voiced, - continuant] and /s/, a voiceless alveolar fricative, is [- voiced, + continuant]. Minimal pairs and sets also occur in sign languages; signs may contrast by hand configuration, place of articulation, or movement.

Some sounds differ phonetically but are non-phonemic because they are in free variation, which means that either sound may occur in the identical environment without changing the meaning of the word. The glottal stop [ʔ] in English is in free variation with the [t] in words such as don't or bottle and is therefore not a phoneme in English.

Phonetic features which are predictable are non-distinctive and redundant. The nasality of vowels in English is a redundant feature since all vowels are nasalised before nasal consonants. One can thus predict the + or - value of this feature in vowels. A feature may therefore be distinctive in one class of sounds and non-distinctive in another. Nasality is distinctive foe English consonants, but non-distinctive and predictable for English vowels.

Phonetic feature which are non-distinctive in one language may be distinctive in another. Aspiration is distinctive in Thai and non-distinctive in English; both aspirated voiceless stops and unaspirated  voiceless stops are phonemes in Thai.

The phonology of a language also includes constraints on the sequences of phonemes in the language, as exemplified by the fact that in English two stop consonants may not occur together at the beginning of a word; similarly, the final sound of the word sing, the velar nasal, never occurs word initially. These sequential constraints determine what are possible but non-occurring words in a language, and what phonetic strings are 'impossible' or 'illegal'. For example, blick [blɪk] is not now an English word but it could become one, whereas kbli [kbli] or ngos [ŋɒs] could not. These possible but non-occurring words constitute accidental gaps.

Words in some languages may also be phonemically distinguished by prosodic or suprasegmental features, such as pitch, stress, and segment duration or length. Languages in which syllables or words are contrasted by pitch are called tone languages. Intonation languages may use pitch variations to distinguish meanings of phrases and sentances.

In English, words and phrases may be differentiated by stress, as in the contrast between the noun pe`rvert in which the first syllable is stressed, and the verb perve`rt in which the final syllable is stressed. In the compound noun ho`tdog versus the adjective + noun phrase hot do`g, the former is stressed on hot, the latter on dog.

Vowel length and consonant length may be phonemic features. Both are contrastive in many dialect of English.

The relationship between the phonemic representation of words and sentences and the phonetic representation (the pronunciation of these words and sentences) is determined by general phonological rules.

Phonological rules in a grammar apply to phonetic strings and alter them in various ways to derive their phonetic pronunciation :
1. They may be assimilation rules that change feature values of segments, thus spreading phonetic properties. The rule that nasalises vowels in English before nasal consonants is such a rule.

2 They may be dissimilation rules that change feature values to make two phonemes in a string more dissimilar like the Latin liquid rule.

3. They may add non-distinctive features that are predictable from the context. The rule that aspirates voiceless stops at the beginning of words and syllables in English is such a rule, since aspiration is a non-phonemic, non-distinctive, and predictable redundant feature.

4. They may insert segments that are not present in the phonemic string. Insertion is also called epenthesis. The historical rule in Spanish that inserted an [e] before word-initial /s/ consonant clusters is an example of an addition or insertion rule.

5. They may delete phonemic segments in certain contexts. Contraction rules in English are deletion rules.

6. They may transpose or move segments in a streing. These metathesis rules occur in many languages, such as Hebrew. The rule in certain dialects of English that changes an /sk/ to [ks] in final position is also a metathesis rule.

Phonological rules often refer to entire classes of sounds rather than to individual sounds. These are natural classes, characterised by the phonetic properties of features that pertain to all the members of each class, such as voiced, or, using + and - signs, the class specified as [+ voiced]. A natural class can be defined by fewer features than required to distinguish a member of that class. Natural classes reflect the ways in which we articulate sounds, or, in some cases, the acoustic characteristic of sounds. The occurrence of such classes, therefore, does not have to be learned in the same way as groups of sounds that are not phonetically similar. Natural classes provide explanations for the occurrence of many phonological rules.

In the writing rules, linguists use formal notations, which often reveal linguistic generalisations of phonological process.

A morpheme may have different phonetic representations; these are determined by the morphophonemics and phonological rules of the language. Thus the regular plural morpheme is phonologically [z] or [s] or [əz], depending on the final phoneme of the noun to which it is attached.

There is a methodology that linguists can use to discover the phonemes of a language, which, includes looking for minimal pairs and complementary distribution. The allophone of a phoneme that results in the simplest statement of the rules of distribution is selected as the underlying phoneme from which the phonetic allophones are derived. The underlying phoneme is selected from the allophone of that phoneme which give the simplest statement of the rule distribution, and the other allophones are derived from it via phonological rules.

The phonological and morphophonemic rules in a language show that the phonemic shape of words or phrases is not identical with their phonetic form. The phonemes are not the actual phonetic sounds, but are abstract mental constructs that are realised as sounds by the operation of rules such as those described above. Non-one teaches us these rules. And yet all speakers of a language know the phonology of their language better than any linguist who tries to describe it. The linguist's job is to make explicit what we know unconsciously about the sound pattern of our language.

More sequential constraints

Some of sequential constarains on phonemes that were discussed previously may show up as phonological and morphophonemic rules. The English homorganic nasal constraint applies between some morphemes as well as within a morpheme. The negative prefix in-, which (like un-) means ‘not’, has three allomorphs which respond to this constraint :

 

The pronunciation of this moepheme is often revealed by the spelling as im- when it is prefixed to morphemes beginning with /p/ or /b/. Because we have no letter ‘ŋ’ in our alphabet (although it exists in alphabets used in other languages), the velar [ŋ] is written as n in words such as incomplete. You may not realise that you pronounce the n in inconceivable, inglorious, incongruous, and other such word as [ŋ] because your homorganic nasal rule is as unconscious, and other rules in your grammar. It is the job of linguists and phoneticians to bring such rules to consciousness or to reveal them as part of the grammar. If you say these words in normal tempo without pausing after the in-, you should feel the back of your tongue rise to touch the velum.

In Akan the negative morpheme also has three nasal allomorphs: [m] before /p/, [n] before /t/, and [ŋ] before /k/, as is shown in the following cases :

We see, then, that one morpheme may have different phonetic forms or allomorphs. We have also seen that one morpheme may occur in the language with the same meaning but different forms in-,un-, and not (all meaning ‘not’). It is not possible to predict which of these forms will occur, so they are separate synonymous morphemes. It is only when the phonetic form is predictable by general rule that we find different phonetic forms of a single morpheme.

The nasal homorganic rule is a feature-changing rule. It can be stated simply as :

Change the place of articulation of a nasal consonant so that it agrees with the place feature value of a following consonant.

In other words, nasal consonants agree in place of articulation with a following consonant.

Given this rule, we can represent this in- negative prefix morpheme by the phoneme representation /ɪn/. Before vowels and before morphemes beginning with /t/ or /d/, the homorganic nasal rule will change nothing since the rule is not violated. The morpheme will be represented by the allomorph [ɪ~n] (after the vowel nasalisation rule applies) as, for example, in the words indecision, interminable, inoperative. Before a morpheme beginning with a labial consonant /b/ or /p/, the alveolar feature of /n/ will be changed by the rule to agree with the place of articulation of the labials, as in impossible and impertinent. Similarly, this feature-changing rule will assimilate the /n/ in /ɪn-/ to a velar nasal before morphemes beginning with /k/ or /g/ in words such as incoherent.

Deriving all forms of the morpheme from /ɪn-/ is the simplest way of revealing this mophological/phonological knowledge. One could represent the morpheme as /ɪm-/ or /ɪŋ-/ instead. But the rule would then have to be complicated.

Change the pace of articulation of a nasal consonant so that it agrees with the place feature value of a following consonant and change the nasal consonant to [n] before a vowel.

This will derive the correct forms of the morpheme but in a more complex fashion than is needed. Rule statements should be as simple and elegant as possible. This principle, known as Occam’s Razor, applies not just in phonology but in all of science. In essence, the simpler the rule, the more general the explanation.

Thus, when two allophones can be derived from one phoneme, one selects as the underlaying segment the allophone that makes the rules and the phonemic feature matrices as simple as possible. For example, deriving the unaspirated and aspirated voiceless labial stops in English from an underlaying /p/ makes aspiration redundant and unnecessary as a phonemic feature value. If /ph/ were the phoneme, the phonemic features would be more complex.

In some cases, different phonetic forms of the same morpheme may be derived by segment-deletion rules, as in following examples :

In none of the words in column A is there a phonemic [g], but in each corresponding word in column B a [g] occurs. Our knowledge of English phonology accounts for these phonetic differences. The ‘[g]-no[g]’ alternation is regular, and we apply it to words that we never have heard before. Suppose someone says :

‘He was a salignant [səlɪgnə~nt] man.’

Even if you do not know what the word means, you migh ask (perhaps to hide your ignorance):

‘Why, did he salign [səlãɪn] somebody ?’ 

It is highly doubtful that a speaker of English would pronounce the verb from with the –ant dropped as [səlɪgn], because the phonological rules of English would ‘delete’ the /g/ when it occured in this context. This rule migh be stated as :

Delete a /g/ when it occurs before a final nasal consonant.

The rule is even more general, as shown by the pair gnostic [nɒstɪk] and agnostic [ægnɒstɪk] and the word cognition, recognition, agnosia, and others, all of which contain the same morpheme related to knowledge. It can be stated as :

Delete a /g/ when it occurs word initially before a nasal consonant or before a word-final nasal.

Given this rule, the phonemic representation of the stems in sign/signature, design/designation, resign/resignation diaphragm/diaphragmatic will include a phonemic /g/ that will be deleted by the regular rule if a suffix is not added. By stating the class of sounds that follow the /g/ (nasal consonants) rather than any specific nasal consonant, the rule deletes the /g/ before both /m/ and /n/.

An alternative analysis is to represent the root morpheme sign as /saɪn/. No /g/ would have to be deleted to derive the verb, but to derive the noun signature an insertion rule would be required and all the words that have a [g] in the derived words an no [g] in the roots would have to be listed. By representing the root morphemes with a phonemic /g/, the regular, automatic, non-excepyional rule of /g/ delection stated above derives the correct forms and also reveals this phonotatic constraint in the language.

The phonological rules that delete whole segments, add segments and features, and change features also account for the various phonetic forms of some morphemes. This point can be further illustrated by the following words :

A speaker of English knows when to pronounce a final /b/ and when not to. The relationship between pronunciation of the A words and their B counterparts is regular and can be accounted for by the following rule :

Delete a word-final /b/ when it occur after an /m/.

Notice that the underlying phonemic representation of the A and B stems is the same.

The rules that delete the segments discussed above are general phonological rules, but their application to phonemic representations results in deriving different phonetic forms of the same morpheme.

Natural classes

Suppose you were writing a grammar of English and wished to include all the generalities that children acquire about the set of phonemes and their allophones. One way of showing what speakers of the language know about the predictable aspects of speech is to include these generalities as phonological rules in the phonological component of the grammar. These are not the rules that someone teaches you at school or that you must obey because someone insists on it; they are rules that are known unconsciously and that express the phonological regularities of the language.
 

In English phonology, such rules determine the conditions under which vowels are nasalised or voiceless stops are aspirated. They are general rules, applying not to a single sound but to classes of sounds. They also apply to all the words in the vocabulary of the language, and they even apply to nonsense words that are not in the language but could enter the language (such as sint, peeg, or sprag, which would be /sɪnt/, /pig/, and /spag/ phonemically and [si`nt], [phig], and [spag] phonetically).

There are also less general rules found in all languages and there may also be exceptions to these general rules. But what is of greater interest is that the more we examine the phonologies of the many thousands of languages of the world, the more we find similar phonological rules that apply to the same board general classes of sounds, like the one we have mentioned nasals, voiceless stpos, alveolars, labials, and so on.

For example, many languages of world include the rule that nasalises vowels before nasal consonants. One need not include a list of the individual sounds to which the rule applies or the sounds which result from its application. We can state the rule as :

Nasalise a vowel when it is followed by a nasal consonant in the sanme syllable.

This rule will apply to all vowel phonemes when they occur in a context before any segment marked [+ nasal], and will add the feature [+ nasal] to the feature matrix of the vowels.

Another rule that occurs frequently in the world’s languages changes the place of articulation of nasal consonants to the place of articulation of the following consonant. Thus, an /n/ will become an [m] before before a /p/ or /b/ and will become a velar [ŋ] before a /k/ or /g/. When two segments agree in their place of articulation they are called homorganic consonants. This homorganic nasal rules occur in Akan as well as English and many other languages.

Many languages have rules which refer to [+ voiced] and [- voiced] sounds. Note that the aspiration rule in English applies to the class of voiceless stops. As in the vowel nasality rule, we did not list the individual segments in the rule since it applies to all the voiceless stops /p, t, k/, as well as to /tʃ/.

That we find such similar rules which apply to the same classes of sounds across languages is not surprising since such rules often have phonetic explanations and these classes of sounds are defined by phonetic features. For this reason, such classes are called natural classes of speech sounds. A natural class is a group of sounds that share one or more distinctive features.

Children find it easier to learn a rule (or construct one) that applies to a natural class of sounds; they do not have to remember the individual sounds, simply the features that these sounds share.

This fact about phonological rules and natural classes illustrates why individual phonemis segments are better regarded as combinations or complexes of features than as indissolube whole segments. If such segments are not be revealed. It would appear that it should be just as easy for a child to learn a rule such as

1. Nasalise vowels before /p/, /i/, or /z/

As to learn a rule such as

2. Nasalise vowels before /m/, /n/, or /ŋ/

Rule (1) has no phonetic explanation whereas rule (2) does. It is easier to raise the velum to produce a nasalised vowel in anticipation of a following nasal consonant than to prevent the velum from rising before the consonant closure.

A natural classes is a set of phonemes that can be defineed by fewer features than any of its members. The classes that include the phoneme /p, t, k, b, g, m, n, ŋ, tʃ, dʒ/ can be defined specifying one feature, [ - continuant]. The phoneme /p/ requires three feature specifications, as does any of the other phonemes in this set.

A class of sounds that can be defined by fewer features than another class of sounds is clearly more general. Thus, the class of [ - continuant, - nasal] sounds in some sense more natural than the class that includes all the segments in that class. Try to do this with feature notation: you will see why such a class is far from natural.

This does not mean that no language has a rule which applies to single sound or even to a class of stops excluding /p/. One does find complex rules in languages including rules that apply to an individual member of a class, but rules pretaining to natural classes occur more frequently, and an explanation is provided for this fact by reference to phonetic properties.

A phonological segment may be a member of a number of classes; for example, /s/ is a member of the class of [+ obstruent]s, [consonants]s, [+ alveolar]s, [+ coronal]s, [- stop]s, [+ continuant]s, [+ sibiliant]s, and so on.