Optimality Theory and Features

3 Optimality Theory and Features * Douglas Pulleyblank

This introduction to featural patterns in Optimality Theory begins by identifying the fundamental function of speech sounds, their role in providing access to meaning. The simplest and most effective way of identifying meanings is to have a unique and invariant phonological address for every distinct meaning. Hearing a particular set of sounds from some language would invariably direct a speaker of that language to a particular meaning. But such a simple mapping of sounds onto meanings is not what we observe in natural language. The complexity that is actually found in sound systems results from the need to reconcile the demands of conflicting constraints on the organization of speech sounds. This chapter focuses on identifying some of these constraints, and showing how Optimality Theory can resolve conflicts between them.

1 Decomposing Sounds into Features It is a standard observation that sentences are composed of words, that words are composed of smaller meaningful units referred to as morphemes (talent, talent+ed, un+talent+ed; judge, judge+ment, judge+ment+al; art, art+ist, art+ist+ic), and that morphemes themselves are composed of strings of sounds; see also Chapter 4. Compare the word a with a single sound, at with two, pat with three, spat with four, and so on.

•Thanks to Darin Howe and Myles Leitch for comments on a draft of this chapter,

as well as to participants in a Holland Institute of Generative Linguistics seminar at the University of Amsterdam for their questions and comments on sections presented there. In addition, a special acknowledgment is due to Diana Archangeli and Terry Langendoen for extremely detailed comments both on the chapter's form and on its substance. The work was supported by Social Sciences & Humanities Research Council of Canada, grant #410-94-0035, and completed under the ausoices of a erant from the Nederlandse oreanisatie voor wetenschaooeliik onderzoek.

60

DOUGLASPULLEYBLANK

However, the decomposition does not stop there. Speech sounds are themselves composed of smaller elements called features. Compare, for example, the consonants found in the following words: pin, zip, bin, man, fin, vet, dad, kin, sit, gas. Although each of these words is composed of three sounds or segments, the features that make up the various sounds crosscut the different consonants. For example, the consonants [p,b,f,v,m] are all labial, articulated with the lower lip; [t,d,s,z,n] are all coronal (or alveolar), articulated by raising the tongue tip or blade to the alveolar ridge, the bony ridge immediately behind the upper teeth; [k,g] are dorsal (or velar), produced by raising the tongue body (or dorsum) to the soft palate (also called the velum). The classification of such consonants according to distinctions based on these and other places of articulation plays a role in numerous phonological patterns. (3.1)

Classification based on place of articulation Dorsal: [k, g, ... ] Labial: [p, b, f, v, m, ... ]

~

/\

pin zip bin man fin vet dad kin sit gas

Coronal: [t, d, s, z, n, ... ] The same consonants form different sets if classified according to different featural dimensions. For example, the consonants [b,v,m,d,z,n,g] are all voiced, produced with vibrating vocal cords; in contrast, the consonants [p,f,t,s,k] are all voiceless, produced with vocal cords that are sufficiently spread apart to prevent any vibration. (3.2)

Classification based on voicing Voiced: [b, v, m, d, z, n, g, ...]

pin zip bin man fin vet dad kin sit gas

~fa4JGG ,, Voiceless: [p, f, t, s, k, ... ] In a similar fashion, these consonants can be classified according to nasality: oral sounds are produced with all airflow through the oral cavity [p,b,f,v,t,d,s,z,k,g]; nasal sounds are produced with air flow through the nasal cavity [m,n]. Finally, the speech sounds in these words can be classified as to whether the airflow in the oral cavity is completely obstructed and then released (stops) [p,b,m,t,d,n,k,g], or whether airflow is continuous and with friction (fricatives) [f,v,s,z]. Each distinct speech sound combines a different set of such featural specifications. Hence [t] is labial, voiceless, oral, and a fricative; [n] is coronal, voiced, nasal, and a stop; [g] is dorsal, voiced, oral, and a stop; and so on.

OPTIMALITY THEORY AND FEATURES (3.3)

61

Classification based on nasality Nasal: [m, n, ... ]

pin zip bin man fin vet dad kin sit gas

Oral: [p, b, f, v, t, d, s, z, k, g, ... ] (3.4)

Classification based on continuancy

A\Z·::/ }\: [mim]pa *! * [min]ba _J_ :z,.,, :-::: / /

-

>

However, this requirement of the grammar to produce voiceless stops is overridden by the effect of the high-ranking constraint that requires consonant clusters to share a single specification for voicing. The tableau in (3.50) shows the effect of adding ICC[VOICE], the cluster constraint requiring that all consonants in a cluster share a single voicing specification (Section 3 ), to the set of constraints governing voicing. (The constraints governing place of articulation have not been included in (3.50) since the issue of place agreement is orthogonal to that of voicing.) (3.50)

lmbabura Quechua: overriding requirement that obstruents be voiceless ICC[VOICE] OBSNOI FAlTH[VOICE) /nan-ta/ T2 *! fian ta \11 w fian-da



E

22

ft

Regardless of the fact that Oas/VO! rules out voiced stops in general, ICC[VoICE] forces their occurrence after a nasal. (Also, the paradigmatic constraint NASNOI would have to be highly ranked to ensure that the nasal remains voiced.) In conclusion, the distinction between voiced and voiceless stops is allophonic in Imbabura Quechua, meaning that the distribution of the two values is predictable and complementary: a voiced stop occurs after a nasal, a voiceless stop occurs elsewhere. Within Optimality Theory, this type of pattern can be derived by the interaction of three constraints. A feature cooccurrence constraint dominates faithfulness, removing the sound from the language's general inventory; some more specific constraint then forces the feature's appearance in some specific context.

.

OPTIMALITY THEORY AND FEATURES

85

The distribution of a feature Feature F occurs in the inventory: FAITH[F] » *F Neutralization: F occurs in an inventory, but a context-specific

condition overrides general considerations of faithfulness: SPECIAL CONDITION ON F » FAITH(F] » *F Feature F does not occur in the inventory: *F » FAITH[F] Allophony: F does not occur in an inventory, but a context-specific

condition overrides the general prohibition: SPECIAL CONDITION ON F » *F » FAITH[FJ

6 Deviation from Identity: Constraints on Association Up to this point, lexical inputs have been presented as though they always consist of complete speech sounds. For example, the meaning of cat is paired with the string of speech sounds /kret/, where the sound k consists of the various features that make up k, the sound a:: consists of the various features that make up a::, and so on. In many languages, however, there are morphemes whose content may be incompletely specified segments, even single features (for recent accounts and references, see Archangeli & Pulleyblank 1994, and Akinlabi 1995, to appear). For example, in Terena, a Southern Maipuran language of Brazil (Akinlabi to appear), the word for 'brother' is ayo. To form the meaning 'my brother', a single feature is added, the feature of nasality: iiY6. No full segment is added to mean 'my' - nasality represents that meaning. Such cases may involve numerous different features. For example, Chaha, a South Semitic language of Ethiopia, indicates feminine imperatives by the addition of a palatalization feature, and third person masculine objects by the addition of labialization (McCarthy 1983). Tiv, a Niger-Congo language of Nigeria represents the general past tense by the addition of a low tone, and the recent past tense by the addition of a high tone (Amott 1964, Pulleyblank 1986). Igbo, also a Niger-Congo language of Nigeria, represents the possessive marker by a high tone, with the precise realization of the tone varying from dialect to dialect (Welmers 1970, Hyman 1975). To illustrate, consider the Igbo case. In Central Igbo, the words in (3 .5 I) are pronounced as indicated when they occur in isolation; a, e, etc. indicate vowels with a high tone, a, e, etc. indicate vowels with a low tone. In Igbo, meanings are distinguished by virtue of the pitch of a vowel. The use of pitch to contrast different meanings is referred to as tone (as mentioned briefly in Section 3). For example, the meanings of the words akwa, akwa, akwa, and akwa are solely differentiated by their tones: the word akwa has two high tones, akwa has two low tones, akwa has a pattern of low tone followed by high tone, while akwa has high tone followed by low.

.....-

DOUGLASPULLEYBLANK

86 (3.51)

lgbo: words in isolation Word Gloss agba 'jaw' el)We 'monkey' akwa 'crying, akwa 'bed' 'egg' akwa 'cloth' akwa

When the first two words of (3 .51) are combined to form a possessive phrase, 'jaw of monkey', they are put together in order iigbii + eIJWe, and the low tone on the final vowel of iigbii is replaced by a high tone (iigb;f). (3.52)

Central Igbo: possessive phrase agba eIJWe 'jaw of monkey'

Recall from Section 2 that an input representation consists of speech sounds paired with a meaning in the mental lexicon. As such, the input to a phrase such as (3.52) would be as in (3.53), where L represents low tone and Hhigh tone. Note that no phonological segment is associated with the high tone of the possessive morpheme. (3.53)

Central Igbo: input representation Phonological representation: Meaning:

L

H

agba jaw

of

A

L

A

eqwe monkey

If the output for such an input were to completely satisfy faithfulness, then it would be identical to the input. As such, no vowel would actually bear the high tone of the possessive morpheme, and the surface tonal pattern would consist of four vowels, all with low tones. Note that such a faithful output would result in the elimination over time of the high tone possessive marker. Since a child learning the fully faithful variety of lgbo would not encounter any reflex of the high tone of the possessive marker, such a generation of speakers would not learn it. The actual output deviates from such a faithful representation by virtue of the appearance of high tone on the final vowel of the first word in the phrase: (3.54)

Central lgbo: output representation L H L I I /".. agba eIJwe

By virtue of its association to a vowel, the high tone is linked to a vowel, and therefore 10 pronounced. The LINK[TONE] constraint is satisfied: 10

such 'linking' is also referred to as 'parsing' in the optimality literature. The term 'link' is used here to avoid confusion with a variety of other uses of the term 'parse'. Note in particular that

OPTIMALITY THEORY AND FEATURES

(3.55)

87

LINK[TONE] : A tone must be associated with a vowel.

For Igbo, associating the high tone of the possessive is more important than being entirely faithful, that is, LINK [TONE] outranks FAITH. (3.56)

Igbo: LINK[TONE] »FAITH

The effect of this ranking is illustrated in the tableau in (3.57). In the first candidate,

*agba CIJWe, the high tone of the possessive marker is not Jinked; in the second and winning candidate, agba CIJWC, the high tone is linked, even at the expense of a faithfulness violation. (3.57)

Central lgbo: the necessity of associating a high tone LINK [TONE] FAITH tagba + H + erJwe/ *! . } _2 ..__