Divided into two parts, Shaping Phonology first explores the elaboration of abstract domains (or units of analysis) that fall under the purview of phonology. These chapters reveal the increasing multidimensionality of phonological representation through such analytical approaches as autosegmental phonology and feature geometry. The second part looks at how the advent of machine learning and computational technologies has allowed for the analysis of larger and larger phonological data sets, prompting a shift from using key examples to demonstrate that a particular generalization is universal to striving for statistical generalizations across large corpora of relevant data. Now fundamental components of the phonologist’s tool kit, these two shifts have inspired a rethinking of just what it means to do linguistics.
|Publisher:||University of Chicago Press|
|Sold by:||Barnes & Noble|
|File size:||17 MB|
|Note:||This product may take a few minutes to download.|
About the Author
Read an Excerpt
The Secret History of Prosodic and Autosegmental Phonology
1.1. Autosegmental Phonology, 1971–1976
The Sound Pattern of English left prosody on the "to do" list as an item for future work (Chomsky and Halle 1968: 329). One problem for such an unshakably segmental formalism was almost immediately evident: although in some languages prosodic features such as tone may be squeezed into a segmental formalism, there are many cases in which "segmental tone features cannot explain and can only awkwardly characterize the operation of tone," such as floating tones, contour tones, word games that manipulate tone independently of segments, and so on (Leben 1973: 25). Some aspects of prosody just are not segmental, by which I mean they are not simply attributes of consonants and vowels, but they relate to other kinds of phonological units. Leben (1973: 42–43) attacked such problems within the framework of generative phonology (as it was then), arguing that, in some languages, tone and nasality were better analyzed as if they were features of morphemes on a separate level of representation from segments:
We may picture the phonological specification of Mende kenya in the following way, where the segmental features are designated schematically as F, F, ..., F and their values are assigned arbitrarily, and where the suprasegmental feature [H] stands for "high-toned":
Leben (1973: 44) went on to define how the tonal morpheme gets mapped onto the segments, from which it is initially independent; for this he introduced a more parsimonious notation in which the originally separate tonal morpheme is written as an initial superscript, and tone features mapped onto segments are written on a second line beneath the vowels and consonants:
(24) TONE MAPPING (Mende)
a. If the number of level tones in the pattern is equal to or less than the number of vowels in the word possessing the pattern, put the first tone on the first vowel, the second on the second, and so on; any remaining vowels receive a copy of the last tone in the pattern.
b. If the number of level tones in the pattern is greater than the number of vowels in the word possessing the pattern, put the first tone on the first vowel, the second on the second, and so on; remaining tones are expressed as a sequence on the last vowel available.
This idea of tone mapping as a mechanism for assigning suprasegmental features to segments was also used by Williams (1971), though Williams and Leben proposed — in the lexicon, at least — separate tonal and segmental tiers, with rules that map the suprasegmental tones onto segments, thus maintaining a rigidly segmental output.
Leben's thesis is dated September 1973. In November 1973, a critical response by Goldsmith (1973) gets rid of this tone mapping operation:
I will suggest that the two tiers — the phonemic and the tonemic — remain separate throughout the segmental derivation. In place of a tone-mapping rule which merges the two tiers into one, I suggest a correspondence rule....
What I would like to suggest to replace a rule of Tone Mapping is a two-tiered approach which retains the two sequences of segments, tonemic and phonemic. ... First I should say that I consider the arguments for suprasegmental tone to be not only interesting but persuasive. The arguments that tones become segmental are, however, very poor. Rather than to say that the toneme becomes mapped into a tone feature on the phonemic segment, I suggest that a correspondence be formally set up between successive tonemes in the tonemic sequence and successive syllabic segments in the phonemic....
There are two alternative and equivalent representations for this system. The first is a graphic one showing the system as literally two-tiered, representing the correspondences with lines. Thus would be represented:
Goldsmith's, Leben's, and Williams's proposals helped to bring suprasegmental phenomena into the arena of generative phonology, providing formalizations that then yielded further insights such as the parallel treatments of subsegmental structure: contour tones on short vowels and contour segments such as short diphthongs and affricates. Goldsmith saw that this theory of representation of tone — autosegmental phonology — immediately found application to a variety of quite distinct and not obviously suprasegmental phenomena, such as short diphthongs, contour segments, gemination, vowel harmony, and nonconcatenative morphology.
A notation for complex segments utilizing multicolumn matrices of distinctive features had been proposed by Hoard (1967) and taken up by Campbell (1974), Anderson (1976), and a few others. For the affricate c, Campbell (1974: 60) wrote:
for the labialized labial pw:
and for the palatalized "palatal" y:
In such examples, the brackets are intended to delimit a single segment, each containing an earlier part (features in the left-hand column) and a later part (features in the right-hand column). Anderson (1976: 335), in a paper that is primarily concerned with the representation of prenasalized and postnasalized consonants, discussed contour tones in similar terms, though he did not explicitly present a complex segment representation for them: "When systems of contour tones are considered in the most general case, and particularly in a number of languages of Africa, it can be seen that tone specifications must in fact have a domain smaller than a single segment. ... Contour tones can occur distinctively on single-mora or short vowels; thus it must be the case that a tone specification in such a case describes only a proper subpart of a single segmental unit."
Goldsmith (1976: 21) presented a possible instantiation of how the complex segment formalism could be used for contour tones on a single segment:
Or one could attempt an equally radical revision of the notion of segment with the introduction of a notation as in (8).
Although the logic of the possibility is clear, this has something of the look of a straw man; for example, no specific citation is given, nor is there one in a manuscript antecedent of this paragraph (Goldsmith 1974: 3):
Some people have suggested breaking up the segment into two parts, and specifying tone features for each half — the first half H, the second L. All the other features of the segment don't get split up, but just belong to the whole segment. This is usually written something like:
With the arrival of autosegmental phonology, complex segment notation persisted for only a short time. The argument was won: the explanatory potential of autosegmental phonology was so far reaching that forty years later, autosegmental representation remains at the heart of the phonology syllabus (e.g., Gussenhoven and Jacobs 2011). Of course, autosegmental phonology did not arise in a vacuum: as Goldsmith knows and has written about himself, he built upon a wide range of earlier work on prosodic features, such as Zellig Harris's long components (Harris 1944) and J. R. Firth's Prosodic Analysis (Palmer 1970). In the remainder of this chapter, I shall try to draw together some perhaps surprising historiographical marginalia on prosodic thinking in the works of our intellectual predecessors. Rather than proceeding strictly chronologically, I shall examine some key periods. In section 1.2, I consider the traditional descriptive approach to prosody in (mainly) the 1920s and 1930s. Then in section 1.3, I trace how J. R. Firth's more phonological-theoretical approach to prosodic phenomena arose in the wake of (post)-Saussurean linguistics, and in particular the importance of Prague School thinking to the shaping of Firthian phonology.
1.2. Tone and Intonation in London, 1920s–1930s
The idea or intuition that suprasegmental units such as tone behave as if they were on a different tier or level from the vowels and consonants and their features has been a common conception for hundreds of years. Steele (1775), for example, gave numerous transcriptions of the intonation and rhythm of speech, such as figure 1.1.
Even at the start of the twentieth century, the treatment of certain aspects of speech — such as tone, intonation, loudness, accentuation, and stress — were recognized as being somewhat problematic for a strictly segmental approach to phonology, as such features are often produced on different time scales from, and have linguistic functions independent of, the "phoneme-sized units," vowels, and consonants.
The problems of analysis presented by suprasegmental features were keenly felt in Daniel Jones's Department of Phonetics at University College, London. At that time, Great Britain had the largest empire the world has ever known, governing massive populations speaking a huge range of languages, in Africa, the Middle East, the Indian Empire (including modern-day Pakistan, India, Bangladesh, and Burma), Malaysia, and Hong Kong. Consequently, linguistic analysis, documentation, and training (of civil servants, for instance) in a wide range of indigenous languages was necessary. Furthermore, international students from across the empire came to Britain — and to London and Oxford, especially — providing linguists such as Jones, Firth, and their colleagues with no shortage of data from African and Oriental languages.
Jones (1918: 137) gave numerous examples of two kinds of intonation transcriptions — linear and musical — with phonemic transcriptions on a separate line (e.g., figure 1.2). An earlier, longer work (Jones 1909) drew the linear pitch traces on a musical stave. The fine detail of these transcriptions is no conceit: even though it must have been a slow and onerous procedure to make them by listening to gramophone recordings, one word or short phrase at a time, Jones had "perfect pitch" perception. Ashby (2015) presented a comparison of Jones's pitch diagrams with f0 analyzed from the original recordings, demonstrating an extraordinary accuracy in Jones's impressionistic transcription (figure 1.3). In Ashby's comparison figure, the measured plot is time-warped at the bar-lines by simple linear extension or compression for comparability with Jones's version, but the pitch in semitones is not manipulated at all.
Phonetic pitch records are not, however, phonological representations. The abstraction over many particular instances to a small number of tune types is seen in later editions of Jones's (1918/1932) An Outline of English Phonetics, in which he follows two of his students, Armstrong and Ward (1926), in recognizing two intonational tunes in English: Tune 1 (early peak, slowly descending to a final brisk fall), and Tune 2 (early peak, similar to Tune 1 to begin with, but ending on a low rise). Jones (1932: 279) noted: "These tunes may be spread over a large number of syllables, or they may be compressed into smaller spaces. ... When the tunes are applied to small groups of syllables or to the extreme case of monosyllables, several of these features disappear." See figure 1.4 for an example.
In terms similar to those of Goldsmith (1978), we might represent Tune 1 in autosegmental notation as H* M* L (or following Pierrehumbert 1981, as perhaps H* L%, with the final fall being from a downstepped H*), and Tune 2 as H* L* M (or H* L*+H%), as in figure 1.5.
The idea of parallel representations for segments and tones, as well as conventions for mapping a shorter sequence of tones onto a possibly longer sequence of syllables already had some currency in descriptive linguistics, especially in relation to African tone languages. Westermann and Ward (1933: chap. 24) provided an extraordinarily modern treatment of tone, including the following remarks:
"The tones of a language must be considered in relation to syllables" (141).
"Every syllable not provided with a tone mark has the tone of the last syllable which has a tone mark [i.e., left-to-right spreading]" (145).
"In any diacritic method of marking tones, the number of marks required is one less than the number of essential tones, since one can be left unmarked: the mid tone is the one usually chosen to be left unmarked" (146).
"Vowel elision is extremely common in African languages and its influence on tones must be considered.... We find that generally speaking the tone of the elided vowel is not dropped entirely, but is often combined with that of the vowel which is retained, and forms a new tone. This new tone keeps some characteristic of both tones. This if the vowel in one of two neighbouring syllables bearing the tones high–low is elided, the resulting syllable has a falling tone" (149).
Ida Ward had moved from Jones's University College, London Phonetics Department to the University of London School of Oriental and African Studies (SOAS) in 1932, becoming head of the African Department in 1937. Many of the departments at SOAS were organized on a geographical basis, and so linguists were engaged there for their expertise in the relevant language areas. After J. R. Firth's migration to SOAS in 1938, and the establishment of the Department of Linguistics and Phonetics in 1940, other key figures of the "Firthian school" were recruited for their expertise in specific language areas. Eugénie Henderson, Lecturer in Phonetics from 1942, for South-East Asian languages such as Thai, Lao, and Cambodian; Jack Carnochan (also Lecturer in Phonetics from 1945) for West African languages such as Hausa, Igbo, and Yoruba; and Keith Sprigg (appointed in 1947) for Tibetan. Firth's own primary language focus was Indian languages — including, given the British organization of the Indian Empire — Burmese.
In many of those Asian languages, lexical tone is often characterized by particular phonatory and/or durational features, as well as pitch patterns; sometimes, the phonatory characteristics may be more reliable or salient cues to a word's lexical tone than its pitch movements. For example, Firth (1933) described Burmese Toneme I (which has slightly falling pitch) as having "vowels of medium length" and creaky voice, when in open or nasal-final syllables, and "very short 'bright' vowels" if closed with a glottal stop. Toneme II has a low level tone, a "gradual ending," and breathy voice. Toneme III also has breathy voice and a "fade-out" ending, but is falling, so in this respect it is distinguished from Toneme III by a different pitch movement. Firth and Rogers (1937) described Hunanese tone 1 as "Long — longer than 3. Normal voice quality. Sometimes ends with slight creak." Tone 2 is "Long — often longer than 1. Voice quality breathy, hollow, 'chesty' with slight initial friction." Though they also note the different pitch movements associated with these tones, they transcribe the tonal distinctions with letters that owe more to their associated phonation, length, and vowel qualities than pitch: tone 1 is marked with syllable-final y or w, depending on the front or back quality of the vowel; tone 2, being breathy, is marked with h; tone 5, being long but not so breathy, is marked by vowel doubling, and tone 4, being short, is marked by single letter vowel spellings.
Sprigg (1955: 125) stated that in Lhasa Tibetan, which has only a two-term contrast, the exponents that have been used as
criteria for setting up the Tonal system are of four orders:–
A. features of pitch.
B. features of duration of vowel.
C. word-initial features.
D. features of voice quality.
For example, Tone One is associated with word-initial voicelessness, whereas Tone Two is associated with word-initial voicing, and may have greater duration and more peripheral vowel qualities. Such a view of tone, which does not privilege pitch, may seem odd to readers of modern phonological handbooks, but it is quite consistent with the traditional Chinese view of tone: for example, the Middle Chinese rù (entering) tone is still marked by short vowels checked by final glottalized stops in modern Cantonese. Another striking example is afforded by Vietnamese, which alongside higher and lower level tones, and higher and lower rising tones, also has higher versus lower rising tones with a glottal stop voice break that divides the rising vowel into two vocalic portions, as in [ba?á] 'residue', [bà?ak] 'silver', in which [a?á] and [à?a] are minimally contrastive, single phonological vowels: higher rising and lower rising , respectively.(Continues…)
Excerpted from "Shaping Phonology"
Copyright © 2018 The University of Chicago.
Excerpted by permission of The University of Chicago Press.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
Table of ContentsPreface
Part One Autosegmental Phonology
1 The Secret History of Prosodic and Autosegmental Phonology
2 A Brief History of Autosegmental Phonology with Special Reference to Its Founder
William R. Leben
B. Applications to Tone
3 The Autosegmental Approach to Tone in Lusoga
Larry M. Hyman
4 Tonal Melodies in the Logoori Verb
C. Extensions of the Theory
5 Autosegments Are Not Just Features
D. Robert Ladd
6 The Importance of Autosegmental Representations for Sign Language Phonology
7 Abstract Underlying Representations in Prosodic Structure
Bert Vaux and Bridget D. Samuels
8 Sonority Waves in Syllabification
9 Toward Progress in Theories of Language Sound Structure
Part Two Computation and Unsupervised Learning
10 On the Discovery Procedure
Jackson L. Lee
11 Model Selection and Phonological Argumentation
James Kirby and Morgan Sonderegger
12 The Dynamics of Prominence Profiles: From Local Computation to Global Patterns
Khalil Iskarous and Louis Goldstein
13 French Liaison in the Light of Corpus Phonology: From Lexical Information to Patterns of Usage Variation
Bernard Laks, Basilio Calderone, and Chiara Celata
14 A Phonological Approach to the Unsupervised Learning of Root-and-Pattern Morphology