Phonology: the Function and Patterning of Sounds

three

PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS

Michael Dobrovolsky

A person's tongue is a twisty thing, there are plenty of words there of every kind, and the range of words is wide, and their variation.

- HOMER, The Iliad, 20

We saw in Chapter 2 that there are a large number of speech sounds that human beings can produce and perceive. No human language exploits all of these possibil- ities. Furthermore, the sounds of all languages are patterned and organized in such ways that linguists can discover some system that underlies their appearance. Linguists generally assume that speakers have (at least) some subconscious knowledge of this system. For example, we saw in Chapter 1 that English speakers recog- nize that forms like slish and screnk are acceptable, while forms like srish and screpk are not. But speakers can do more than note that certain forms are unnatural in their system; they can also correct such forms to make them acceptable in their own language. Without knowing exactly why, most English speakers would pronounce a form like srish as [sari/] rather than saying [nj] or [si/]. It is this largely subconscious knowledge of sound patterns that phonologists are interested in uncovering. This chapter is about phonology, the component of a grammar made up of the elements and principles that determine how sounds vary and pattern in a language. Phonologists attempt to make explicit formal statements about the sound patterns of individual languages in order to discover something about the linguistic knowledge that people must have in order to use these patterns. Even more broadly, the study of phonology attempts to discover general principles that underlie the patterning of sounds in human language. The existence of patterns in language depends on the organization of certain basic elements or units that combine to make up these patterns. Three major units of analysis will be presented in this chapter. We are already acquainted with the idea that the flow of speech can be divided into segments. In this chapter, we investigate 64 CHAPTER THREE

the patterned variation of segments. A second unit of phonological analysis is the feature. Features may be thought of as the smallest building blocks of phonological structure, corresponding as they do to articulatory or acoustic categories such as [voice] or [strident]. We will also learn how segments are built up from features and how this fact is connected with phonological patterning. We will also investigate the properties of syllables in phonology. A syllable is a unit of linguistic structure that consists of a syllabic element—usually a vowel—and any segments that are associ- ated with it. (The word segment itself can be divided into two syllables: seg and ment.) These units—feature, segment, and syllable—interact with the processes we investi- gated in the previous chapter and with certain general principles to produce the sound patterns of language. The existence of patterned elements implies that there must be some principles of organization. Features, segments, and syllables are organized into hierarchical lev- els; each level is composed of elements at the level beneath it. In Figure 3.1, the word-level unit segment is represented by the abbreviation Wd. This word consists of two syllables, each of which is represented with the Greek letter a (sigma). Each syllable is made up of a number of segments (the internal structure of syllables is also hierarchical, and is treated in Section 5 of this chapter). Each segment, in turn, is composed of features. A representation of segment is given in Figure 3.1. (For pur- poses of illustration, only a few features are provided for each segment.)

Wd Word level

Syllable level

Segment level

-syllabic +syllabic -syllabic -syllabic +syllabic -syllabic -syllabic Feature level -sonorant +sonorant -sonorant +sonorant +sonorant +sonorant -sonorant •9 '• +coronal -high +high +labial +reduced +nasal +coronal

Figure 3.1 Partial phonological representation of segment Us ) In the next sections, we look at the phonological knowledge that enables speakers to distinguish among forms and to deal with the considerable phonetic variation Si found in the pronunciation of speech segments.

£..;'•

E:, 1 SEGMENTS IN CONTRAST

All speakers know which segments of their language contrast. Segments are said to Ill: contrast (or to be distinctive or be in opposition) when their presence alone may distinguish forms with different meanings from each other. The segments [s] and [z] contrast in the words sip and zip, as do the vowels of hit, hate, and hot.

lit:: PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 65

1.1 MINIMAL PAIRS A minimal pair consists of two forms with distinct meanings that differ by only one segment found in the same position in each form. The examples [sip] sip and [zipj zip given previously form a minimal pair and show that the sounds [s] and [z] contrast in English. A number of minimal pairs that demonstrate consonant contrasts for English are given in Table 3.1; remember it is on the basis of sound and not spelling that minimal pairs are established. In displaying contrasts, contrasting words are often placed along the horizontal axis with respect to their place of articulation, reading from left to right (labial, alveolar, and so on), and vertically with respect to manner of articulation, in order to show which places and manners of articulation are exploited by the language in question.

Table 3.1 Contrasts among consonants in English Labial Interdental Alveolar Alveopalatal Velar Glottal

Stops and affricates (noncontinuants) tap [p] pat [t] match ItJl pick [k] tab |b] pad [d] Madge Pig Igl

Continuants leaf [f] thigh [9] sip [s| mesher in hip [h] . leave [v] thy [5] zip fz] measure bi

Nasals

sum [m] sun fn] sung [rjj

Liquids and glides wet [w] yet lil leer [1] rear [r]

The phonetic context in which a sound occurs is called its environment. Pairs that show segments in nearly identical environments, such as azure/assure or author/either, are called near-minimal pairs. They help to establish contrasts where no minimal pairs can be found. You may assume that two segments contrast once a minimal pair or a near- minimal pair has been established. It is in fact rare to find minimal pairs for all distinctive sounds in all environments in a language, since the historical evolution of every language has led to some sounds being utilized more frequently than others, or being eliminated from some environments. For example, you will find no minimal pairs involving [h] and [n] in word-initial or word-final position in English, because there are no words that begin with [rj] or end in [h]. It is also difficult to find 66 CHAPTER THREE

minimal pairs in English that have the phone [3], which occurs for the most part in words borrowed from French such as azure and mirage. Vowel contrasts in English Contrasts among English vowels can be established with a few sets of examples. For now, we will continue to assume that English vowel-glide sequences like [ij], [uw], [ow], and so on are single vowels. From this perspective, we can say that the vowels [ij] and [1], [ej[ and [e], and so on, contrast.

Table 3.2 Vowel contrasts in American English

beet Ibijt] [ij] bit [bit] [i] bait [bejt] [ejl bet [bet] [el bat [bat] l*\ cooed [k"uwdi [uw] 2 P could [k'-ud] [0] V code [khowd| [owj A caught [k»3t] t>] cot [k"at] [a] cut [k"At] [A| lewd [luwd] [uwj loud [lawd| [aw] lied [Iajd] [aj] Lloyd [lojd| [ojl

1.2 LANGUAGE-SPECIFIC CONTRASTS Contrasts are language-specific; sounds that are distinctive in one language will not necessarily be distinctive in another. For example, the difference between the two 2.1 vowels [e] and [ae] is crucial to English, as we can see from minimal pairs like Ben [ben] and ban [baen]. But in Turkish, this difference in pronunciation is not distinctive. A Turkish speaker may pronounce the word for 'I' as [ben] or [baen], and it will make no difference to the meaning.

Table 3.3 Language-specific vowel contrasts: English versus Turkish

English Turkish

[ben] Ben [benl [baen] ban [baen]

Conversely, sounds that do not contrast in English, such as long and short vowels, may be distinctive in another language. There are no minimal pairs of the type [haet]:[hae:t] or [luws]:[lu:ws] in English. But in Japanese and Finnish, short and long vowels contrast, as the next examples show. PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 67

Table 3.4 Short/long vowel contrasts in Japanese and Finnish

Japanese

[tori] 'bird' [tori:] 'shrine gate' [kiboj 'scale' [kibo:] 'hope'

Finnish

[tuli] 'fire' [tu:li] 'wind' [haeta?] 'distress' [ha?:tae:] 'to evict'

Establishing the contrasting segments in a language is a first step in phonological analysis. But in any language, there are many sounds that never contrast. The following section deals with this aspect of phonological analysis.

PHONETICALLY CONDITIONED VARIATION: PHONEMES AND ALLOPHONES

Everyday speech contains a great deal of phonetic variation that speakers pay little or no attention to. Some of this variation arises from nonlinguistic factors such as fatigue, excitement, orthodontic work, gum chewing, and the like. This kind of variation is not part of the domain of phonology. But much phonetic variation is systematic. It occurs most often among phonetically similar segments and is conditioned by the phonetic context or environment in which the segments are found. This variation occurs because segments are affected and altered by the phonetic characteristics of neighboring elements or the larger phonological context in which they occur. We rarely notice this kind of variation because every speaker has the ability to factor it out in order to focus attention on only the relevant contrasts of the language.

2.1 COMPLEMENTARY DISTRIBUTION When first learning phonetic transcription, English speakers are often surprised that all the Is they pronounce are not identical. In Table 3.5, the /s in column A are voiced, while those in column B are voiceless (indicated here by a subscript o). Many speakers of English are unaware that they routinely produce this difference in articulation, which can be heard clearly when the words in column B are pronounced slowly.

Table 3.5 Voiced and voiceless / in English

A B

blue [bluw] plow [plaw] gleam [glijm] clap [klip] slip [slip] clear [k!ijr] flog [flag] play [Plej] leaf [lijf] 68 CHAPTER THREE

The voicelessness of the Is in column B is a consequence of their phonetic environment. Voiced and voiceless Is vary systematically in that all of the voiceless [I]s occur no predictably after the class of voiceless stops. Since no voiced (1] ever occurs in the wit same phonetic environment as a voiceless one (and vice versa), we say that the two So variants of / are in complementary distribution. At Table 3.6 Complementary distribution of [IJ and [1] in English mi ant HI Ian After voiceless stops no yes alsi Elsewhere yes no tio: we Ian The term elsewhere is used in Table 3.6 to indicate the wider distribution (occurrence \ in a greater number of different phonetic environments) of voiced [1]. It occurs after .-] voiced stops, voiceless fricatives, and in word-initial position. \ lisl In spite of these phonetic differences, native speakers consider the two English Is \ get to be instances of the same segment, since they are phonetically similar and the dif- J we ferences between them are systematic and predictable. This perception of sameness I ilk is supported by the fact that the two /s never contrast in English. There are no min- 1 imal pairs like [plejj and [plej]. We can sum up the relationship that the two Is bear 1 to each other by stating that, for speakers of English, the two Is are phonetically dif- 1 ferent but in the sound system of English, given their phonetic similarity, pre- j dictable distribution, and noncontrastiveness, they are plionologically the same. j -\ We -•] foL 2.2 PHONEMES AND ALLOPHONES I The ability to group phonetically different sounds together into one class is shared by all speakers of all languages. This phonological knowledge is represented formally on a level of phonological representation that is distinct from phonetic representation. Predictable variants of certain segments are grouped together into a contrastive phonological unit called a phoneme. These variants, which are referred to as allo- Th' phones, are usually phonetically similar and are frequently found in complemen- COl tary distribution. A representation of this relationship is shown in Figure 3.2. The are phonemic symbol for the class—generally the same symbol as the elsewhere vari- clu ant—is placed between slashes, and the symbols for allophones are enclosed in pho- dis netic brackets. an<

(Phoneme —> ) III Ian SOI (Allophones —>) fl] [1] w am After voiceless stops Elsewhere but wh Figure 3.2 The phoneme /I/ and its allophones [1] and [1| in English m voi Allophonic variation is found throughout language. In fact, every speech sound j Soi we utter is an allophone of some phoneme and can be grouped together with other q ful phonetically similar sounds into a class that is represented by a phoneme on a pho- g I

.'38 PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 69

nological level of representation. An important part of phonological analysis deals with discovering the phonemes of languages and accounting for allophonic variation. Some problematic distributions At this point, some other considerations in determining phonemes and allophones must be taken into account. So far, we have seen that a minimal pair test is a quick and direct way of establishing that two sounds belong to separate phonemes in a language. If the sounds contrast, they are members of different phonemes. We have also seen that if certain sounds are noncontrastive and in complementary distribution, they may be considered allophones of one phoneme. In some cases, however, we must go beyond these procedures to discover the phonological inventory of a language. As noted in Section 1.1, certain patterns of distribution prevent some sounds in a language from ever contrasting with each other. In cases like these, we can establish the phonemic status of a sound by default. If the sound cannot be grouped together with any other phonetically similar sounds as an allophone of a phoneme, we may assume it has phonemic status. The following data from English help to illustrate this point.

1) *[rjowp] (does not exist) [howp] 'hope' •faejt] (does not exist) [hejt] 'hate' We can see here that [h] and [rj] do not contrast in initial position in English. The following examples show that they do not contrast in final position either. 2) [Ian] 'long' *[lah] (does not exist) [sin] 'sing' *[sih] (does not exist) [kterj] 'clang' *[klaeh] (does not exist) I These lists could be extended for pages, but a minimal pair involving [h] and [rj] | could never be found in English. Additionally, as 1) and 2) have shown, [h] and [rj] I are in complementary distribution. Do these facts taken together not lead us to con- f elude that [h] and [n] are allophones of one phoneme? No. Since [h] and [rj] are so I distinct phonetically, we assume that each one is a member of a separate phoneme I and that the pattern of distribution is of secondary importance in this instance. I Minimal pairs or near-minimal pairs help us establish which sounds contrast in a | language; phonetic similarity and complementary distribution help us decide which sounds are allophones of a particular phoneme. But not all examples of variation I among sounds can be dealt with through these approaches. i In some cases, phonetically similar sounds are neither in complementary distri- 1 bution nor are they found to contrast. It is still possible, nevertheless, to determine ! which phonemes these sounds belong to. A case in point is the variation in English voiceless stops when they are found in word-final position, as in the word stop. Sometimes an English speaker releases the articulation of these sounds rather force- fully. Let us represent this with a diacritic sign [!]. At other times, the same speaker may keep the articulators closed for a moment after the articulation; the diacritic f] 70 CHAPTER THREE

can represent this. Some speakers may even coarticulate a glottal closure (represented here with the raised symbol for a glottal stop following the consonant in question) and produce the word as [stap?]. Thus we can find at least three pronunciations of stop: [stap!], [stop"1], and [stap7]. Since there is no difference in the meaning of these forms and since the final consonants are phonetically similar, we say that these sounds are in free variation, and that they are all allophones of the phoneme /p/. The same pattern holds for the other voiceless stops of English.

2.3 CLASSES AND GENERALIZATION IN PHONOLOGY Phonological analysis permits us to account for the great amount of phonetic variation in everyday speech. This systematic variation is widely extended within languages. Compare the English data in Table 3.5 with those in Table 3.7.

Table 3.7 Voiced and voiceless allophones of English /r/ A B brew [bruw] prow [praw] green Igrijn] trip [tnpl drip [drip] creep Ikrijpl frog [frag] pray [PFJl shrimp Lfnmp]

The data show that the allophones of English /r/ pattern like those of English /I/. Based on this information, we can state that there is an /r/ phoneme in English with (at least) two allophones—one voiced, the other voiceless. But if we were to stop there, we would overlook an important point. The phonemes /r/ and /I/ belong to the same class of sounds: both are liquids. By taking this information into account, we can state a general fact about English.

3) In English, liquids show voiceless allophones after voiceless stops and voiced allophones elsewhere. A major goal of phonological description is the formulation of the most general statements possible about sound patterns. Reference to classes of segments helps accomplish this. Additional data from English illustrate this point.

Table 3.8 Voiced and voiceless allophones of English glides

A B beauty [bjuwrij] putrid [pjuwtnd] Duane [dwejn] twin [twin] Gwen [gwsn] quick [kwikl view [vjuw] cute [kjuwt] swim [swim] thwack [Bwaek] PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 71

These forms demonstrate that the contrasting glides /j/ and /w/ each pattern like the liquids. We can now extend our general statement even further. 4) In English, liquids and glides have voiceless allophones after voiceless stops, and voiced allophones elsewhere. Clearly, allophones do not pattern piecemeal, but rather according to their mem- bership in phonetic classes.

2.4 CANADIAN RAISING In most Canadian and some American dialects, pronunciations like those illustrated in Table 3.9 are common.

Table 3.9 Low and central vowel allophones in raising dialects

In Table 3.9, the vowels [aj] and [AJ] are in complementary distribution. The [aj] occurs before the class of voiced consonants or in word-final position, and the [AJ] occurs before the class of voiceless consonants. The two are allophones of a single phoneme /aj/. The same relationship holds between the vowels [aw] and [AW], which are allophones of /aw/.

/aj/ /aw/

[aj] [Aj] [aw] [AW]

Before a voiced Before a voiceless Before a voiced Before a voiceless consonant or in consonant consonant or in consonant word-final position word-final position 1 j «: Figure 3.3 Allophones of /aj/ and /aw/ in raising dialects •4 Bs Again, we see this phonological fact reflected in everyday language use. Most speakers of these dialects find it difficult to distinguish between these allophones, even when the difference is pointed out to them. This is because the difference is not I 1: contrastive. On the other hand, many people who speak varieties of English that do not have the [AJ] or [AW] allophones are very much aware of their presence in Canadian English. To them, a Canadian speaker sounds markedly different, even though they may be confused about the nature of the difference. 72 CHAPTER THREE

I don't agree he was an American.... Where all other English-speaking people pronounce OU as a diphthong, the Canadian .. . makes a separate sound for each letter. The word about, for instance, he pronounces as ab-oh-oot.

Philip MacDonald, The List of Adrian Messenger

This phenomenon is sometimes referred to as Canadian Raising, since the more re- stricted allophones [AJ] and [AW] have higher vowel components than the elsewhere allophones [aj] and [aw].

2.5 ENGLISH VOWELS AND GLIDES A final example of predictable variation that refers to classes of segments is again taken from English. Table 3.2 showed contrasts among English vowels. In English, the nonlow, tense vowels [ij], [ej], [uw], and [ow], always contain one of the two glides [j] and [w]. Note that the labiovelar glide [w] occurs with the back rounded vowels [u] and [o] and the palatal glide [j] occurs with the front unrounded vowels [i] and [ej. Stated in other words, we can say that the back rounded vowels predictably co-occur with the back (labiovelar) glide and the front vowels co-occur with the nonback unrounded glide. These facts are summed up in Table 3.10.

Table 3.10 Tense vowel-glide combinations in English

Vowel Glide Vowel Glide (both nonback and unrounded) (both back and rounded) i j [fijt] feat u w [buwt] boot e j [fejt] fate - o w [bowtj boat

These data show parallels with the allophonic distribution we have considered so far: certain elements are predictable under certain systematically stateable phonetic conditions. Here, however, instead of a number of variants of a phoneme, we have two segments whose distribution is predictable: these glides are always found after nonlow tense vowels in English. This predictability is analogous to the predictability of allophonic variants—if certain sounds are predictably found in a given environment, they are not included in the phonemic representation. We can thus draw from the data in Table 3.10 the following generalization about the English vowel system. 5) The nonlow tense vowels of English are predictably followed by a glide that has the same backness and roundedness as the vowel. Given the variation in English vowels that we have examined, we are now able to summarize what we have discovered about the English vowel system. The English vowel system We have already seen that certain elements of phonemes, such as voicing, vowel height, etc., are not present in phonological representations because they are pre- PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 73

dictable from the phonetic context in which the allophone is found. Since the presence of the appropriate glides following the class of nonlow tense vowels is predictable, they need not be present in the phonological representation of the English vowels either. These generalizations enable us to represent the contrasts between words like heat and hit, late and let, cooed and could purely as tense versus lax vowel contrasts, and not as contrasts that also involve the presence of glides. We can also represent the vowel found in words like code and snow without a following glide. Figure 3.4 sums up the differences between phonetic and phonological representations for the vowels of English. heat [ill N cooed [uw] /u/ hit [i] III could lu/ late [ej] lei luck [A] M/ code [ow] lol let [e] lei caught [3] hi sat [*] Is/ cot [a] /a/ boy [oj] /oj tide [aj] /aj/ loud [aw] /aw/ Figure 3.4 Phonetic and phonological representations of English vowels

2.6 LANGUAGE-SPECIFIC PATTERNS Although the phenomenon of allophonic variation is universal, the patterning of phonemes and allophones is language-specific. What we discover for one language may not hold true for another. Language-specific variation in allophonic nasalization It is not unusual for nasal vowel allophones to occur near a nasal consonant, but, as Table 3.11 shows, the patterning may vary from language to language.

Table 3.11 Nasal vowels in Scots Gaelic 'big' % [mo:r] 4 [nl] 'cattle' [ne:l] 'cloud' [mu] 'about' [ru:n] 'secret'

Scots Gaelic has oral and nasal vowel allophones. Here we can state the following.

6) Vowels are nasal in Scots Gaelic when preceded or followed by a nasal consonant. Malay, a language spoken in Malaysia and Singapore, presents another variation on the theme of nasal allophones. 74 CHAPTER THREE

Table 3.12 Nasalization in Malay [me wan] 'luxurious' [majarj] 'stalk' [marah] 'scold' [nae?] 'ascend' [mSlararj] 'forbid' [makan] 'eat' [rumah] 'house' [kareta] 'car'

Here, all vowels and glides following a nasal are predictably nasalized until an obstru- ent, liquid, or glottal ([h], [?]) is reached. For Malay the generalization is as follows. 7) In Malay, all vowels and glides following a nasal consonant and not separated from it by a non-nasal consonant are nasalized. Language-specific variation in allophonic distribution As was shown in Section 1.2, a phonemic contrast in one language may not prove to be a phonemic contrast in another. This means that the relationship of phonemes to allophones may vary. A comparison of the contrasts among stops in English and Khmer (Cambodian) illustrates this point. In both languages, aspirated and unaspirated phones can be heard.

Table 3.13 Stop phones in English and Khmer

English Khmer

[Pi [ph] [Pi [ph] W [th] [t] [th] [k] [kh] [k] [kh]

In English, aspirated and unaspirated stops are allophones of their respective phonemes (the distribution is explained in Section 4.5 of this chapter); there are no contrasting forms like [prk] and [phik]. In Khmer, unaspirated and aspirated voiceless stops contrast.

Table 3.14 Khmer contrastive voiceless stops

h [PD:rj] 'to wish' [p J:rj] 'also' h [tDP] 'to support' [t op] 'be suffocated' [kat] 'to cut' [khat] 'to polish'

The phonological contrasts of the two languages are different, even though the phones are not. These distributions are the same for the other voiceless stops in both languages. PHONOLOGY: THE FUNCTION AND PATTERNING OF SOUNDS 75

English Khmer

(Phonemes) /p/ /p/ /p'V

(Allophones) [p] [p»] [pj [phJ Figure 3.5 English and Khmer voiceless bilabial stop phonemes and allophones

3 PHONETIC AND PHONEMIC TRANSCRIPTION

Having seen how nondistinctive properties of segments are factored out by phonological analysis, we can now compare the type of transcription used for segmental phonological representation with phonetic transcription. The following examples show this difference for the classes of sounds in English that we have examined so far.

Table 3.15 Phonetic and phonemic transcription Predictable property(s) not Phonetic Phonemic represented in phonemic transcription transcription Word transcription [plaw] /plaw/ plow voicelessness of liquid [krijpj /krip/ creep voicelessness of liquid; glide after nonlow tense vowel [kwikj /kwik/ quick voicelessness of liquid [lejt] /let/ late glide after nonlow tense vowel [let] /let/ let — [thajdj /tajd/ tied aspiration

[fAjt] /tajt/ tight aspiration; Canadian Raising

The contrast between phonetic and phonemic representation is even more strik- ing for the Malay forms given earlier, as is shown in Table 3.16.

Table 3.16 Phonetic and phonemic transcription of Malay nasal vowels Predictable property(s) not Phonetic Phonemic represented in phonemic transcription transcription Word transcription

fmewah] /mewah/ 'luxurious' nasalization [majan] /majan/ 'stalk' nasalization [nag?] /nae?/ 'ascend' nasalization

Here, nasalization on all vowel and glide segments is predictable and is therefore omitted from the phonological representation.