| This is the print version of Phonology
You won't see this message or any elements not part of the book's content when you print or preview this page.
What is 'phonology'?[edit | edit source]
Phonology is the study of the mental organization of speech sounds in language, and its relation to other language behaviors.
Languages can differ in the ways that the sound system is organized. For example, languages may differ radically in how stress is assigned. In English, stress may distinguish the meaning of a word (compare record, record), and its location must be memorized as part of the word. However, in Choctaw, stress is often completely predictable: every even-numbered syllable will be stressed (except the last syllable of a word) in specific, but frequent environments. One goal of phonology is to describe the stress systems of different languages. More generally, one goal of phonology is descriptive: to document the kinds of sound patterns that occur in different languages.
Besides describing what occurs in different languages, linguistic theory is also concerned with explanation: why do certain patterns occur over and over again in many different languages, while others appear to be completely impossible, and still others appear to be possible but very rare? Taking stress again as an example, phonologists have documented the following generalizations:
- in many languages, stress always falls in the same position in a word, which is called predictable stress
- if stress is predictable, it always falls on one of the first two syllables, or one of the last three
There is no language, for example, which requires stress to fall on the middle syllable of a word. An explanatory theory of stress should tell us why there are no languages of this type, as well as giving insight into why particular languages behave the way they do. In fact, this question makes up one of the subfields of phonology known as accentual typology.
Another subfield of phonology deals with the issue of what sequences of consonants and vowels are allowed. This is the study of phonotactics -- constraints on what sound sequences are allowed in the words of a language. Example (1), below, illustrates that English and Spanish have different phonotactics.
|word-medial||ostrich, huskey, raspy||estado, escultor, espam|
|word-initial||state, sculptor, spam||--ungrammatical--|
|word-final||past, ask, ga
English has many words beginning with consonant clusters of the form /sC/ (where C is any consonant). Spanish does not and cannot have words of this form. What Spanish does have are closely related words (known as cognates) which begin with /esC/ where English has /sC/. These facts suggests that Spanish has a phonotactic restriction against syllables that begin with /sC/. This interpretation makes sense out of the fact that when native Spanish speakers learn English as an adult, they sometimes insert an 'extra' vowel (e.g. speed --> espeed). This mispronunciation makes the English form consistent with the phonotactics of Spanish.
Another major topic is alternations. An alternation occurs when the 'same' word or sound sequence is pronounced differently, depending on the context. For example, in American English, the 't' is pronounced differently in atom than in atomic:
- atom [ˈæɾəm]
- atomic [əˈtʰɐmɪk]
The 't' in 'atom' -- technically an alveolar tap, [ɾ] -- is very weak; in fact, most American English speakers cannot hear the difference between this 't' and the 'd' in 'Adam'. In contrast, the 't' sound in atomic is very strong; it is pronounced with a brief puff of air called aspirations -- technically this sound is an aspirated alveolar stop, [tʰ]. Thus, the 't' in atom alternates between [ɾ] (atom) and [tʰ] (atom-ic). This kind of alternation is especially interesting because [ɾ] and [tʰ] are phonetically very different, and yet most native English speakers do not even notice the difference. But the difference is obvious to native speakers of other languages; for example, the [ɾ] in atom sounds like an 'r' to native Japanese speakers, while the [tʰ] in atomic sounds like an unusually strong 't'.
There are several other major topics in phonology, including prosody and intonation, vowel (and consonant) harmony, process interaction (and opacity), computational phonology, historical phonology, and others. The examples of phonotactics and alternation were chosen to illustrate central and interesting phenomena that phonology seeks to explain.
Core assumptions[edit | edit source]
All generative theories assume a fundamental division of labor between information that is stored, and information that is computed dynamically, known as the lexicon and the grammar, respectively:
- lexicon -- mental component that stores arbitrary, unpredictable aspects of the pronunciations of words and other sequences
- grammar -- mental component that enforces systematic aspects of pronunciation that hold throughout a language
The distinction between grammar and lexicon is motivated by the fact that some aspects of pronunciation are systematic throughout an entire language, while others vary idiosyncratically from word to word. An example of the former is that in English, voiceless stops are always aspirated (produced with a puff of air) when they occur immediately before a stressed vowel, e.g. 'cat' [kʰæt]. This property holds throughout the language, rather than varying between words, so it is systematic. Because this generalization holds throughout the entire language, it is best explained by the assumption that speakers possess unconscious generalizations as to licit sound sequences in their language -- a phonological grammar. An additional, interesting fact is that these generalizations can vary from language to language. For example, voiceless stops are not aspirated in Japanese or Spanish. This tells us that some aspects of a language's phonological grammar must be learned (because an infant who is exposed to Japanese learns not to aspirate voiceless stops, while an infant exposed to English learns to aspirate voiceless stops before stressed vowels).
An example of a lexical property is the pronunciation of an individual form. For example, the generic form for a domestic feline (cat) begins with a voiceless velar stop (/k/) in English. In Japanese, the generic form for a domestic feline (neko) begins with a denti-alveolar nasal /n/. Not all words begin with a /k/ in English, or with an /n/ in Japanese. Therefore, the initial consonant of a word is specific to the word, and must be stored in the lexicon.
Overview[edit | edit source]
Phonetics refers to the physical production of speech sounds, the acoustic properties of speech sounds, and certain aspects of how speech sounds are perceived. Many researchers contrast phonetics with phonology, where phonology is supposed to deal with the mental representation of speech sounds, and the grammar which maps lexical representations to phonetic categories.
The chapter is organized into four parts. Part I covers the anatomy and physiology of the vocal tract -- the names for the parts. Part II covers articulation -- how speech sounds are produced. Part III covers acoustics -- properties of the sound waves that are generated by different kinds of speech sounds. Part IV covers perception -- how listeners recover the speech sounds that were produced from sound waves. The remainder of this overview gives an extremely brief overview of how speech works.
In vowels, the larynx is a 'source' which converts air pressure into a series of soud waves, generating the 'pitch'. The upper vocal tract shapes and molds the sound originating from the larynx, causing some integer multiples of the fundamental frequency ('harmonics') to resonate ('formants'). Vowel quality is primarily determined by how wide open the mouth is (jaw height), whether the tongue comes closer to the roof of the mouth at the hard palate (front) or soft palate (back), and whether the lips are rounded. In consonants, a sharp constrictipon of airflow creates a pressure differential in the upper vocal tract. Consonants are named by the articulator at which the primary constriction occurs, as well as how tight the constriction is, and the status of the larynx. For example, /p/ is called a voiceless labial stop because the primary consonantal constriction occurs at the lips (labial), the lips are completely closed so they completely stop all airflow, and the larynx is not generating modal voicing.
Manner of articulation refers to how tight the oral constriction is. For example, in the word pen the initial sound /p/ consists of a complete blockage of airflow at the lips, whereas in the word when the initial sound /w/ consists of only a very partial blockage of airflow at the lips. Place of articulation refers to where the primary consonantal constriction occurs. For example, /p/ is produced at the lips while /t/ is produced just behind the teeth. Voicing refers to whether the larynx is generating modal voicing during a consonant. For example, /s/ as in hiss is voiceless, while /z/ as in his is voiced. (You can hear the difference between voiced and voiceless sounds by putting your finger on your throat right in front of the Adam's apple and saying both; for example say hissssssss and hizzzzzzz to feel the voicing contrast between /s/ and /z/ at the end of a word.)
Sonorants -- formants. Obstruents -- turbulence and VOT
Part I: Anatomy[edit | edit source]
Anatomy and Physiology of the Vocal Tract[edit | edit source]
- The Larynx (voice box)
- The Tongue
- The Lips
- The Teeth
- The Roof of the Mouth (palate)
- Other Parts
Part II: Articulation[edit | edit source]
Phonation[edit | edit source]
- Modal Phonation
- Creaky Voice (aka Laryngealization, Vocal Fry)
- Breathy Voice
The Production of Vowels[edit | edit source]
- Vowel Height
- Tongue Frontness/Backness
- Lip Rounding
- Other Vowel Features
Consonants[edit | edit source]
- Place of Articulation
- Manner of Articulation
- Voicing and Other Laryngeal Features
- Other Consonantal Features
The International Phonetic Alphabet[edit | edit source]
Part III: Acoustics[edit | edit source]
Part IV: Perception[edit | edit source]
Autosegmental Phonology[edit | edit source]
Cantonese: Lexical Tone, 1-1 Association, Allotony[edit | edit source]
Mende: Lexical Tone, Stacking/Spreading[edit | edit source]
Japanese: Pitch accents[edit | edit source]
Tiv (example write-up): When lexical and grammatical tones compete[edit | edit source]
In Tiv, verbal stems exhibit tonal alternations as a function of grammatical tense. The tonal alternations are schematized below using six stems (va 'come', dza 'go', uⁿgʷa 'hear', veⁿde 'refuse', jevese 'flee', and ⁿgohoro 'accept') conjugated in three tenses: general past (1), past habitual (2), and recent past (3).
(1) General past (GEN)
|number of sylables||Class I examples||Class II examples|
(2) Past habitual (HAB)
|number of sylables||Class I examples||Class II examples|
(3) Recent past (REC)
|number of sylables||Class I examples||Class II examples|
It turns out that all monosyllabic verb stems exhibit the same pattern of alternations as either /va/ 'come' or /dza/ 'go'. Similarly, all disyllabic verb stems exhibit the same pattern of alternations as either /uⁿgʷa/ 'hear' or /veⁿde/ 'refuse', and all trisyllabic verb stems exhibit the same pattern of alternations as either /jevese/ 'flee' or /ⁿgohoro/ 'accept'. Thus, the data in (1-3) represent the full range of tonal behaviors that occur in this part of the language. The goal of this section is to analyze the tonal system that is at play in (1-3).
Since tone is the phenomena of interest, and all mono-, di-, and tri-syllabic forms are like (1-3), we will begin by abstracting away from the segmental content, re-representing the data as only tones:
(4) Tonal conjugation in Tiv
|Class I||Class II|
|General past (GEN)||H, HL, HLL||L, LL, LLL|
|Habitual past (HAB)||H, HH, HHH||H, LH, LHH|
|Recent past (REC)||H, HH, HHL||H, LH, LHL|
In the next section, we propose that there are two tonal melodies available for verb stems in this language, and that this is what distinguishes class I from class II. We furthermore propose a default process of left-to-right tone association.
Proposing URs, part I: Lexical tones for class I and II[edit | edit source]
Inspecting (4), we see that every class I form begins with an H tone, while nearly every class II form begins with an L tone:
example 5: tonal possibilities, with lexical tones highlighted
Therefore, we posit that an H melody is part of the lexical representation for all of the class I forms, while an L is part of the lexical representation for all of the class II forms. For simplicity, we will hereafter refer to class I as the H class, and class II as the L class.
This assumption can be captured using an autosegmental representation with one tier for tone and another for segmental material. For example, the URs for va 'come (class I)' and dza 'go (class II)' are given in (6):
example 6: URs for H-class monosyllable come/va and L-class monosyllable go/dza
As a key assumption of autosegmental phonology is that the tonal melody is autonomous from the segmental melody, there must be some way to associate tones to tone-bearing units (vowels). In most languages, the association process is left-to-right, 1-1, and iterative, so we assume the following:
example 7: Association rule
Since (7) applies iteratively from left to right, the effect is to associate the leftmost tone with the leftmost vowel, and then to associate the second tone with the second vowel, and so on, until either tones run out, or vowels run out. Thus, if a Tiv speaker were to produce the bare form of va, we might assume the derivation would go as in (8):
example 8: Derivation for hypothetical bare form for Tiv
In order to do a derivation for a tensed form, we must have the complete UR. As we will see shortly, there is reason to believe that each tense has a UR consisting of a pure tonal melody.
Proposing URs, part II: Grammatical tone for the past tenses[edit | edit source]
In (5), we bolded the initial tone of each form and lightened the remaining tones, which helped to visually highlight the commonalities among all class I and class II forms. From this, we inferred that (in most cases) there is a lexical L or H tone that occurs at the beginning of the form. That is, we can factor this part of the word into a lexical tone. We repeat this process in (9), except that here we gray out the lexical tone, and instead highlight the remainder of the tonal pattern:
example 9: same as (4), but with the grammatical tones bolded
Note that in (9), and hereafter, we have begun to refer to class I as the 'H class' (because we assume these stems contain an underlying/lexical H tone), while we refer to class II as the 'L class' (because we assume these stems contain an underlying/lexical L tone). In addition, we have abbreviated the tenses as gen, hab, and rec, respectively.
After 'factoring out' (greying out) the lexical tone in (9), we observe that there is a fully systematic pattern in the disyllabic and trisyllabic stems. In the general past, the second syllable receives an L tone, and so does the third syllable if there is one. In both the habitual and recent pasts, the second syllable receives an H tone. However the habitual and recent pasts are distinguished on the third syllable (if there is one): habitual past always has a level H pattern (H/HH), while recent past has a falling pattern (H/HL). Therefore, we posit that these tenses have a purely tonal UR:
example 10: grammatical tone in Tiv: URs for general past, habitual past, and recent past
As before, there is some exceptional behavior for the habitual and recent pasts in L class stems (highlighted), which we will ignore for the moment. Instead, we confirm that these URs can explain some of the existing items with derivations.
example 11: derivations of hear/ungwa(H)-gen(L), hear/ungwa(H)-hab(H), refuse/vende(L)-gen(L), and refuse/vende(L)-hab(H)
Example (11) shows derivations for 'hear' (/uⁿgʷa, H/) and 'refuse' (/veⁿde, L/), conjugated in the general past (/, L/) and the habitual past (/, H/). In this case, the tonal patterns are especially simple. The verb stem contributes a lexical tone (H for the H class stem 'hear', L for the L class stem 'refuse'). The verbal inflection contributes a grammatical tone (L for gen, H for hab). Association applies, so that the lexical tone is realized on the first syllable, and the grammatical tone is realized on the second syllable. We can similarly explain the allocation of tones to the trisyllabic forms in the recent past:
example 12: derivations of flee/jevese(H)-rec(HL) and accept/ngohoro(L)-rec(HL)
Taking stock of where we are, we have posited H and L lexical tones for the two classes of verbs in (6), posited tonal melodies for the various past tenses of Tiv in (10), and a default association process which associates unassociated tones to unassociated vowels from left-to-right in (7). With these assumptions alone, we are able to explain 6 out of the 18 tonal patterns that occur in the language -- the easiest cases in which the number of tones (lexical + grammatical) matches the number of tone-bearing units (vowels).
Tonal spreading when there are extra vowels[edit | edit source]
With URs and the default association rule as background, we are prepared to deal with cases where there are more vowels than tones. These occur in the trisyllabic forms for the general and habitual past (jévèsè, jévésé; ngòhòrò, ngòhóró), since there are three syllables but only two tones (one lexical, one grammatical). As with the forms derived earlier in (11-12), we observe that the putative lexical tone occurs on the first syllable, while the grammatical tone appears on both the second and third syllables. That is, we assume the output of the derivation must be a phonological representation like in (13):
example 13: putative SR of accept/ngohoro(L)-hab(H)
The use of multiple association lines emanating from the grammatical H tone indicate that although there is one tone in the UR, it is expressed across multiple vowels in the SR.
The association process that we described before is only supposed to be responsible for associating completely unassociated tone/vowel pairs. As stated, it does not handle cases in which there are extra vowels (or extra tones). Therefore, one option for handling cases like (13) is to modify the Association rule. However, we will pursue an alternative option -- supplying a Spreading rule that applies after Association:
example 14: spreading rule
(Since Spreading is ordered after Association, the relevant tone will always be the last tone in a string.) A derivation for (13) ngòhóró ('accept'-hab) is given in (15):
example 15: derivation of accept/ngohoro(L)-hab(H)
The derivations for jévèsè, jévésé, and ngòhòrò are entirely parallel to (15). Thus, with the lexical and grammatical UR tones, Association rule, and Spreading, we are able to account for 10 out of the 18 tonal patterns shown in (4). Our progress is indicated in (16), with items accounted for indicated with green highlighting. As before, the past habitual and recent past L class monosyllables are exceptional because they do not bear an L tone, and so these are marked with yellow highlighting.
example 16: showing tonal patterns that we have accounted for with green highlighting, unaccounted for as regular text, and exceptions in yellow
When there are too many tones, part I[edit | edit source]
Next, we turn to simple cases in which there are two of the same tone which compete for a single syllable. This occurs in monosyllabic L class in the general past ('go': /dza, L/ + gen: /, L/) and monosyllabic H class in the past habitual ('come': /va, H/ + hab: /, H/). An autosegmental representation of the URs is given in (17):
example 17: URs for come/va(H)-hab(H) and go/dza(L)-gen(L)
In this case, the outcome is indistinguishable, regardless of whether it is the lexical tone or the grammatical tone which ends up associating to the vowel. However, since we do not observe falling or rising tones on individual vowels (despite the opportunity to do so, in the falling tone of the recent past), we may assume that this language enforces a strict requirement that there must be exactly one tone associated to each vowel. In cases like (17), where it does not actually matter which tone wins since they are identical, we posit a process of tonal Simplification. We will assume that Simplification is ordered after Association (so that the leftmost tone is already associated).
example 18: simplifaction rule
The order of Simplification cannot be established with respect to Spreading, since Simplification only applies if there are more tones than vowels, while Spreading only applies if there are more vowels than tones. Since the order does not matter, we will put Simplification before Spreading (alphabetical order). A derivation for 'come'-hab is given in (19):
example 19: derivation of come/va(H)-hab(H)
Note that derivation (19) crucially differs from the hypothetical derivation in (8) (the uninflected form of /va/) in that there is a grammatical tone present in the UR which is deleted owing to Simplification. Thus, while the derived SR is the same, (19) correctly accounts for the inflected (past habitual) form. The derivation for 'go'-gen is entirely parallel, but with low tones. Thus, we are now able to account for 12 out of the 18 possible tonal patterns, as shown in (20).
example 20: record-keeping (like 16)
In the remaining 6 cases, there are more tones than vowels, but not all of the tones are the same.
When there are too many tones, part II[edit | edit source]
We begin by considering /va, H/ in the general past (/, L/). The UR is given in (20):
example 21: UR for come/va(H)-gen(L)
In this case, there are two tones which must compete for the syllable. As evident from the surface form (vá), it is the lexical tone that 'wins' (since otherwise we would expect the grammatical low tone to be expressed in the SR). However, something different occurs with (/dza, L/) in the past habitual (/, H/):
example 22: UR for go/dza(L)-hab(H)
In this case, it is the grammatical tone which is expressed, rather than the lexical tone. We can conclude that it is not lexical/grammatical status that determines the winner when two tones compete. Rather, it must be some property that is shared by the lexical tone in /va, H/ and the grammatical tone of the past habitual (/, H/). One obvious property is that both of these tones are high. Put another way, the thing that is constant between (20) and (21) is that when a low tone and a high tone compete for the same syllable, the high tone wins. This holds true regardless of whether the high tone occurs before the low tone, as in (20), or after it, as in (21).
We will formulate this as two rules. The first rule, called H Greediness, states that an unassociated H tone can replace (eliminate) an immediately preceding associated L tone:
example 23: H Greediness rule (iterative, left-to-right)
The second rule, Stray Erasure, states that an unassociated tone must delete.
example 24: Stray Erasure (iterative, left-to-right)
H Greediness must be ordered before Stray Erasure (otherwise, Stray Erasure would delete all unassociated high tones first, so that H Greediness would never get a chance to apply). In fact, Stray Erasure should be ordered last, since it is essentially a bookeeping/cleanup operation to handle any remaining unassociated tones. In the case of Tiv, the cleanup operation is to delete remaining tones, but we could easily imagine other options; for example in Mende, all remaining unassociated tones are stacked onto the final syllable of the word. As it turns out, the relative ordering of H Greediness and Simplification is not important, so long as Stray Erasure can apply to any tone. (In fact, it would be reasonable to formulate the two as a single rule, since they both deal with a sequence of two tones, of which the first is associated. However we have kept them as two rules here, since this more complex rule is tricky to formalize.) Derivations for 'come'-gen, 'go'-hab, 'hear'-rec, and 'refuse'-rec are given below.
example 25: derivations for come/va(H)-gen(L), go/dza(L)-hab(H), hear/ungwa(H)-rec(HL), and refuse/vende(L)-rec(HL)
It turns out that this rule ordering straightforwardly accounts for the two remaining causes, of monosyllabic stems with the grammatical contour tone (recent past):
example 26: derivations for come/va(H)-rec(HL) and go/dza(L)-rec(HL)
Summary of Tiv[edit | edit source]
This section has treated the tonal system of the past verbal conjugations of Tiv. The facts to be accounted for are the limited number of tonal patterns. More specifically, the tonal pattern on a verb form is predictable from the past tense inflection, the number of syllables, and the verb class. The available set of patterns are shown in (4), repeated here as (27):
example 27: repeat of 4
We accounted for these data by positing an underlying (lexical) H tone in the H class items (e.g. va 'come'), and an underlying/exical L tone in the L class items (e.g. dza 'go'). In addition, we posited grammatical tone (i.e. the phonological spellout of these inflectional morphemes is a pure tone, with no segmental content): L for the general past, H for the past habitual, and HL for the recent past.
In the simplest cases, the surface tone on a form is expressed by the lexical tone on the first syllable, followed by the grammatical tone on the remaining syllables (Association rule). If there are extra vowels which do not yet bear a tone, the final tone spreads rightwards (Spreading rule). Conversely, if there are more tones than vowels, they must compete for the final syllable. Tiv regulates this competition by favoring H over L: an unassociated H can replace an associated L, while an associated H is not replaced by an unassociated L (H Greediness rule). Finally, any remaining unassociated tones are removed by Stray Erasure1. With these assumptions, we can economically explain the tonal alternations in the entire past tense conjugation system of Tiv.
English: Phrasal melodies[edit | edit source]