A National Map of the Regional
Dialects of American English


July 15, 1997

 

A National Map of

The Regional Dialects of American English

William Labov, Sharon Ash and Charles Boberg

 The Linguistics Laboratory,
Department of Linguistics,
University of Pennsylvania

The Telsur Project of the Linguistics Laboratory of the University of Pennsylvania is engaged in a telephone survey of the sound changes affecting the English of North America.[1] The aim of this project is the production of the Phonological Atlas of North America, which will chart the present state of the phonological systems of urban dialects, and the advance of sound changes in progress. A first sample of the urbanized areas of the United States was completed as of June 1, 1997, yielding data on the vowel systems of 607 Telsur speakers. All speakers are local to the speech community studied, raised in that city. For all speakers, phonetic transcriptions of minimal pairs and elicited forms yield information on the progress of ongoing mergers. Maps showing the current state of six mergers on the basis of the main sample are now available on the Phonological Atlas home page. In addition, the Phonological Atlas home page also provides access to the paper on "The organization of dialect diversity in North America." This was presented at the ICSLP4 meeting in Philadelphia in October 1996 and contains the general framework for our approach to sound changes in progress, with maps on the data available at that time.

 

This communication is a first report on the analysis of mergers and chain shifts in progress on the basis of the acoustic analysis of the stressed vowels of 238 speakers. It presents our answer to the two questions:

 

(1) What are the major dialect regions of the United States?

 (2) What are the defining features of those regions?

 

Though the map provided here still represents work in progress, it provides a first national overview of the sound changes affecting American English which is the principal aim of the Telsur project. The current state of American dialects is the result of the active processes of change and diversification that we have been tracing since 1968 (Labov Yaeger and Steiner 1972 [LYS 1972]; Labov 1991, 1994, 1996; Ash 1994). Many scholars believe that there are no discrete dialect boundaries, and that there are no clear and discrete dialect divisions of a language. That is the explicit position of Carver 1987, whose location of American dialect boundaries on the basis of regional vocabulary is not radically different from our own. However, the results to be given here show clear and distinct dialect boundaries, delineating areas with a high degree of internal homogeneity. Moreover, the features defining these reflect the fundamental organizing principles of American dialects and the sound changes that have been transforming those dialects for the past century.

 Map 1 shows four major dialect regions: the Inland North, the South, the West, and the Midland. The first three show a relatively uniform development of the three major sound shifts of American English, each moving in different directions. The fourth area is a residual domain with much greater diversity, where most individual cities have developed dialect patterns of their own.

The data base

The basis for the dialect divisions in Map 1 is a data set created by the impressionistic coding and acoustic analysis of the 238 vowel systems. The total number of tokens analyzed for each system varied from 200 to 900, with a mean of 271 tokens for each vowel system. The number of tokens analyzed for a given phoneme varied from 5 to 50, depending upon the frequency of the vowels and the importance for the process being analyzed. The work of vowel analysis is still in progress for additional speakers. The total number of 607 speakers covers the urbanized areas in a fairly uniform way, with two subjects for each urbanized area, and 4 to 6 for the larger metropolises. In the early stages of the project, the main emphasis was upon the North Central and Midwestern States, with special attention to the North/North Midland boundary. In these areas, vowel systems were analyzed for a number of smaller speech communities, with populations of 20,000 to 50,000, as well as the major metropolises, where data on six speakers was obtained. As the project expanded to a national sample, the process of acoustic analysis did not keep pace with the progress of the interviewing. In particular, the data base of Map 1 is deficient in the following ways:

 a. No data are given for New York City and Philadelphia, whose vowel systems have been extensively reported elsewhere (LYS 1972, Labov 1994).

 b. A minimal number of points are given for Eastern New England.

 c. There is minimal coverage of the South, with only one speaker for most major cities representing the Gulf States, as compared to two or more speakers for the Midland and the North.

 d. There is inadequate coverage of the large populations of California and the Pacific Coast.

 e. There is no attempt to represent the distinct dialects of many small and well-defined rural areas, like Martha's Vineyard, the Iron Range of Michigan, and the Outer Banks of North Carolina.

 f. No data is yet available for Canada, although the interviews have been completed.

 

Within these limitations, Map 1 gives a coherent national picture of the state of the sound changes involved.

 

The data input to the mapping programs is a spreadsheet with 200 columns of data and one row for each speaker. The 200 columns contain 18 items of demographic information, 18 items from the impressionistic coding of Telsur speakers, and 164 items of acoustic information, based the F1 and F2 means of the 32 vowel classes, and statistical comparisons of the means for various means and allophones. 2 These included the distance between means, the significance of the difference based on t-tests, and the angle formed by the vector connecting the means and the F2 axis. A particularly important measure for the progress of certain sound changes is the quadrant into which such an angle falls. Thus for the relation between the means of short /e/ and the nucleus of /ey, there are four values of such a measure:

 1 /e/ is higher and fronter than /æ/

 2 /e/ is higher and backer than /æ/

 3 /e/ is lower and backer than /æ/

 4 /e/ is lower and fronter than /æ/

 

The F1 and F2 values for all vowel systems are normalized, using the log-mean normalization (Nearey 1977), which has proved to be an efficient and reliable way of eliminating differences due to vocal tract length while preserving differences due to social and geographic factors (Hindle 1978).

Dialect regions of the United States

The three major dialect regions of the United States identified in Map 1 -- the Inland North, the South, and the West -- correspond to the three vowel patterns first presented in "The Three Dialects of English" (Labov 1991). They are the major expanding patterns that are actively forming the linguistic landscape of the country., As developed in this paper, the phonological center of these opposing patterns are the Northern Cities Shift in the Inland North, the Southern Shift in the South, and the Low Back Merger in the West. Since that time, a fourth phonological pattern, the Canadian Shift, has been reported in Clark, Elms and Youssef 1995. Discussion of this pattern will be presented when the Canadian interviews are analyzed.

 Map 1 also identifies a number of distinct and important dialect areas in the Eastern United States, which were clearly set out in the work of the Linguistic Atlas (Kurath and McDavid 1961): Eastern New England; New York City; and the Mid-Atlantic coastal area encompassing Philadelphia, Wilmington and Baltimore. No phonological basis for a division between the Upper South, the Lower South, and the Gulf States is presented here, though a separate dialect area is recognized around Charleston and Savannah.

 The great contribution of Kurath to American dialectology, the identification of the Midland region, is well represented in Map 1, but somewhat transformed. A major part of the South Midland--the Appalachian cities of Kentucky, West Virginia and Tennessee--is here rejoined to the South. The Midland region then includes the major cities of Philadelphia, Pittsburgh, Akron, Columbus, Dayton, Indianapolis, Peoria, Evansville, the Quad Cities, St. Louis, Des Moines, Kansas City, Wichita,, Lincoln and Omaha. It can be differentiated from the other regions by general phonological criteria, and it is itself divided into a lower and upper half by a characteristic sound change, the fronting of checked /ow/. But in contrast with the Inland North and the South, no single set of sound changes is identified with the Midland region. On the contrary, the various Midland cities show localized patterns, which are shifting and diverging from each other in many ways. The importance of the Midland region in this report rests not upon the description of a single "Midland" phonology, but rather the fact that the northern and southern boundaries of the Midland turn out to be the discrete and influential boundaries that determine the shape of American dialect geography. The North/North Midland line falls almost exactly where it was first placed on the basis of lexical evidence in Kurath 1949, and further developed by Shuy 1962 and Carver 1987 on the basis of additional lexical markers. Although much of this vocabulary is obsolete or evanescent, and the phonological correlates laid out in Kurath and McDavid 1961 have largely disappeared, the North/North Midland line remains as an almost impermeable boundary to the southern expansion of the Northern Cities Shift.

Much of the controversy surrounding the concept of the Midland conept has rested on evidence for the traditional view that the line between North and South is really the most important division in American English, corresponding to the fact that this is the only distinction that can be reliably identified by the American public (Preston 19??). Bailey 1968 argued that the line running along the Ohio River was more important than the Midland boundaries, on the basis of phonological patterns of syllabification. Carver's national map of dialect divisions, based on the data of the Dictionary of American Regional English, makes the North/South division pre-eminent, and reduces the North/North Midland line to a secondary division between Upper and Lower North. Our own delineation of the boundary of the South coincides closely with that of Carver 1987 from Maryland to the Mississippi River, as shown in Map 2. From that point on, our boundary diverges, and extends further north and further west, to include Arkansas, Southern Missouri, and the four speakers in Texas we have studied.

 A remarkable finding of Map 1 is that the major phonological boundaries of the U.S. as determined by new and vigorous sound changes which arose in the 20th century coincide with the major lexical boundaries based on vocabulary wjocj was largely set in the early settlement periods. Map 2 shows that the coincidence of the North/North Midland boundary with that of Carver 1987 is also quite close. The major differences are the northward expansion of the Pittsburgh area in Map 1 to include Erie, and the more precise definition of the eastern and western boundaries of the Inland North. The implications of this coincidence for the general theory of sound change remain to be explored in future publications.

 The West has long been considered to be a residual category, without any clear character of its own. Its mixed and diffuse character has been attributed to the result of settlement routes in which the influence of Northern, Southern and Midland patterns were intermingled. Although the Telsur project finds some evidence to support this view, Map 1 shows that the West is also the center of its own new and developing phonological system.

The taxonomy of American dialects

The legends of Map 1 identify the major phonological features that define the areas. This small set of sound features were chosen on the basis of two criteria:

 (1) They show maximal geographic clustering indicated by an isogloss with the minimal percent of occurrences outside the area along with the maximal proportion of occurrences inside it.

 (2) They are drawn from the set of systematic sound changes that differentiate the phonology of that area, and are linked causally to that wider set of features as shown by correlations and isogloss bundles.

Many of the distinguishing features on Map 1 include numbers in Hertz or in Cartesian distances between F1/F2 locations. These numbers are the result of the minimax criterion just described. The specific threshhold values are of course subject to change as more data are obtained or as the sound changes proceed. They represent the quantitative implementation of a qualitative distinction such as "fronting of checked /ow/." In further publications of Atlas results to follow, thematic maps will be provided for many features in which the size of the symbol is proportional to the quantity measured.

 The relation of these features to each other and to Map 1 is shown in a first overview by the taxonomy of Figure 1. This figure begins with the vowel configuration known as the "initial position" (Labov 1994:163-166). It represents our best estimation of the common base for American English dialects which resulted from the mixing of various English dialects in the 16th and 17th centuries. The initial position is the point of departure for the sound changes that now differentiate the American dialects on Map 1.[3] The phonological symbols represent the historical word classes given in (1):

 

(1)
 
SHORT VOWELS
LONG VOWELS
Front Upgliding
Back Upgliding
Ingliding
Front
 
High
/i/
bit
/iy/
beat
 
Mid
/e/
bet
/ey/
bait
 
Low
/æ/
bat
/ay/
bite
 
Back
High /u/ put /uw/ boot
Mid /^/ but /oy/ 
/ow/
quoit 
boat
/oh/ bought
Low /o/ pot /aw/ bout
 

In the initial configuration, the long high and mid vowels /iy/, /ey/, /uw/ and /ow/ are generally diphthongal, but the distance between nucleus and glide is quite small, especially in checked position before a consonant. The nuclei are located in the tense, peripheral positions shown in Figure 1, higher and more peripheral than the corresponding short nuclei /i/, /e/, /u/ and /^/.[4]

The North

The first major differentiation of dialects shown in Figure 1 concerns the position of the long high and mid vowels. In the North, these generally retain the initial position, while in the South and Midland, these nuclei are laxed. In the course of the Southern Shift, they become progressively lowered so that high nuclei become mid, and mid nuclei become low.[5] All of the dialects in which this initial position is retained may be grouped within a general region called The North.

The North Central region

The dialect area that best preserves the features of the initial position is the North Central region. On Map 1 the lightly shaded circles indicate the most conservative speakers, for whom the second formant of checked /ow/ is less than 1100 Hz. This back position of /ow/ in boat, road, home, etc. is matched by an equally conservative position of the other /ow/, /uw/ and /ey/ allophones. The same points are selected if we choose as criteria the conservative position of free /ow/ (< 1350 Hz), of free /uw/ (<1850 Hz), and front checked /ey/ (> 1850 Hz).[6]

 Table 1 shows the distribution of the defining features of the North Central and Inland North regions. The back position of checked /ow/ is characteristic of 25 of the 35 points in the North Central region, or 71%. Eighteen speakers with back checked /ow/ lie outside this area, almost all in the neighboring regions of the West, the Midland and the Inland North. Back position of /ow/ is also characteristic of the other conservative area, Western New Englan.d.[7]

 

Table 1
 Distribution of defining features of the North Central and Inland North regions
 
No.Cen.  Inl. 
No. 
No. 
Mid 
So. 
Mid 
West  So.  St. 
 Louis 
Pitts.  Phila  ENE  WNE  Tot 
No. of speakers  35  54  34  37  24  27  238 
F2 /owC/ < 1100  25  33  72 
F2/e/-F2/o/ < 375  47  61 
The light colored circles shown in Map 1 for the Inland North are in a way deceiving: these represent the few conservative speakers of the Inland North who are not strongly involved in the Northern Cities Shift, to be described below. Table 1 shows that 33 of the 54 speakers in the Inland North show the back position of checked /ow/, or about two thirds. In general, the speakers of the Inland North also show the conservative position of the long high and mid vowels that is found in the North Central region. There is no incompatibility between the two criteria, since as we will see, the Northern Cities Shift concerns only short vowels. In other words, the peripheral position of long high and mid vowels is largely shared by the North Central and Inland North regions. Within the two areas, 65% of the speakers share this characteristic (internal consistency); for the country as a whole, 80% of those with this back form of checked /ow/ are located in the North Central and Inland North regions(external consistency). This criterion for defining the North Central region is almost matched in consistency by the front peripheral character of /ey/ and the back peripheral character of free /ow/.

 The conservative character of the North Central region is best exemplified by the speakers of Northern Iowa, who come as close to the initial position as any in the country. The more northerly parts of this region show the well known monophthongal character of the long high and mid vowels, which is not registered in this report. The stereotype of Minnesota speech, for example, is expressed in the pronunciation of Minnesota with a long monophthongal o: {mên'so:t'}.

The Inland North

The Inland North is one of the most populous areas of the United States, a region of large cities surrounding the Great Lakes and spread out through upper New York State. This region is defined by a revolutionary rotation of the English short vowels, which historically have remained stable since the 8th century. This region is defined as a linguistic unit by the uniform involvement in the Northern Cities Shift [NCS] and several related features of the vowel system. The six connected elements of the NCS are shown in (2).

 

(2) The Northern Cities Shift

 .

The NCS begins with the wholesale raising and fronting of the /æ/ phoneme, the triggering event which differentiated the Inland North from New England, as registered on Figure 1. The second stage, the fronting of short /o/, was first noted in Fasold's unpublished report on the movement of /æ/ and /o/ in his Detroit study (1969). LYS 1972 give acoustic and impressionistic evidence for stages 1,2,4 and 6 based on recordings in Chicago, Detroit, Buffalo, Rochester and Syracuse. Stage 5 was first reported by Eckert in the course of her research in the suburbs of Detroit, completing the chain; in further papers Eckert shows how the various stages are related in temporal and social space (1986, 1988, 1989, 1991). The Project on Cross-Dialectal Comprehension at the Linguistics Laboratory investigated further development of the NCS in Chicago, along with effects on communication within and across dialects (Labov 1989, Labov and Ash in press). Labov 1994 relates the NCS to the general theory of chain shifting. Gordon 19?? gives acoustic and social analyses of the current stages of the NCS in various Michigan communities.

 Since the six stages of the NCS are connected in a systematic chain shift, there are many combined movements that may be used as indices of the existence and degree of development of the shift. One of the most effective of these capitalizes on the fact that stages 2 and 4 move the short vowels /e/ and /o/ in opposite directions. As /o/ is fronted and /e/ is backed, the F2 distance between them declines. In a conservative speaker like Tessa C. of Mason City, Iowa, normalized /e/ shows an F2 value of 1821, and /o/ shows 1403. An advanced speaker in the inland North, like Martha F. of Kenosha, Wisconsin, has a normalized F2 value of 1786 for /e/ and 1763 for /o/. (For more details, see Labov 1996). Map 1 therefore identifies all of the NCS speakers as those for whom the difference in the F2 of /e/ and /o/ is less than 375 Hz.

 Speakers identified in this way are shown by dark shaded circles. The line defining the Inland North encloses New York State (Binghamton, Syracuse, Rochester, Buffalo), the northeast portion of Ohio bordering Lake Erie (Cleveland, Akron, Lorain, Elyria), all of Michigan (Detroit, Ann Arbor, Flint, Grand Rapids, Kalamazoo), northwest Indiana bordering on Lake Michigan (Gary), northeastern Illinois (Chicago, Rockford) and southeastern Wisconsin (Kenosha, Madison). One Great Lakes city is excluded from the Inland North area: Erie, which now shows the merger of /o/ and /oh/ and other features characteristic of Pittsburgh and the western Pennsylvania region.

 Table 1 shows that within the Inland North region, 47 of 54 speakers show the configuration defined above. Eight of the ten speakers who do not are marked with the light colored circles that indicate the conservative position of the long high and mid vowels. Outside of the Inland North, we find only 11 speakers with the dark shaded circles that register the approximation of F2 for /e/ and /o/: 5 in the North Central region, 4 in the North Midland, 1 in the South Midland, and 1 in Western New England.[8] This criterion of the NCS shows 87% internal consistency in the Inland North (47/54), and 81% external consistency for the country as a whole (47/58).

 Most importantly, Map 2 indicates that the border separating the North from the North Midland coincides well with the boundary of lexical features in the Dictionary of American Regional English. As future publications of the Atlas will show, this is also the southern boundary of other features of the NCS, of Canadian Raising and the relative back position of /aw/ compared to /ay/. The North/North Midland boundary is therefore one of the most profound divisions in American phonology.

Eastern New England

The present set of Telsur acoustic analyses include only three speakers from Eastern New England: two from Boston and one from Waterville, Maine. The defining characteristics are those which were identified by Kurath and McDavid 1961: the merger of /o/ and /oh/ in cot and caught, Don and dawn, etc., and the vocalization of postvocalic /r/. All three of these speakers show these features, as well as the general northern configuration discussed above. Map 1 of the Phonological Atlas home page shows the /o/~/oh/ data for all 607 speakers interviewed, including 16 speakers from Eastern New England. Eight of these show a complete merger in production and perception, and none show a complete distinction. In comparison, the neighboring city of Providence shows four of six speakers with a complete distinction in production and perception, and no cases of total merger. The other criterion, the vocalization of /r/, is eroding under the influence of the post World War II convention that constricted /r/ is the appropriate standard for careful speech (Labov 1966). However, all three speakers show some vocalization of /r/, and one Bostonian shows 50%. Though most of Western New England shows consistent [r], Providence shares the vocalization of /r/ under the same conditions as the rest of Eastern New England. In this respect, the relations of Eastern New England to its neighbors are stable.

New York City

Given the large body of acoustic data available on New York City speech in previous research (Labov 1966, LYS 1972), no further analyses were conducted by Telsur. Recent observations in New York City indicate that the city vernacular is intact, and the general configuration provided in those publications remains. New York shows the stable peripheral position of the high and mid vowels /iy, ey, uw, ow/ characteristic of the North. The everyday speech of the city exhibits consistent vocalization of postvocalic /r/ except for the mid-central vowel in bird, and when a final /r/ is followed by a vowel in the next word. As indicated in Figure 1, the short a word class is split into lax /æ/ and tense /æh/ sets local to New York City, and /oh/ is raised to mid and high position. As a result, the Cartesian distance between /oh/ and /o/ is maximal, a feature shared with Philadelphia (see below).

Western New England

Map 1 shows Western New England as a residual area, surrounded by the marked patterns of Eastern New England, New York City, and the Inland North. The city of Providence has a unique configuration as noted above, with vocalization of /r/ combined with a stable /o/~/oh/ distinction, but without the vowel raising characteristic of New York City. No clear pattern of sound change emerges from western New England in the Kurath and McDavid materials or in our present limited data. However, it must be remembered that most of the people who settled the Inland North came from this region, and that the generating conditions for the NCS may be found there. In this respect, we note with interest that our Hartford speaker satisfies the NCS criterion.

The South

We now follow the left-hand branches of Figure 1, tracing the dialects which show a laxing and centralization of the long high and mid vowels. The first branch on this left half of the tree is defined by the well known southern feature, the monophthongization of /ay/.[9]

 

Bailey 1997 traces the history of 23 features of white vernacular English in Texas from a variety of sources, using acoustic analysis of early recordings. The monophthongization of /ay/ is shown in his Table 2 to be absent before 1875, variable from 1875 to 1945, and consistent after 1945. The monophthongization of this vowel can be seen as the triggering event in one section of the complex pattern which has been termed "The Southern Shift" (LYS 1972, Labov 1991), shown in (3).

 

(3) The Southern Shift

 

The stages of the Southern Shift are numbered from 1 to 8 for convenience of reference; the temporal sequences are not as well established as for the Northern Cities Shift. The monophthongization of /ay/ is shown as the first event, opening the way for [2]. the complete lowering of the nucleus of /ey/ along a non-peripheral track, which is followed in some areas by [3] a similar movement of /iy/. With stage [4] we have the relative reversing of the position of the nuclei of the long and short vowels, as the short vowel nuclei assume peripheral positions and develop inglides. Stages [5] and [6] involve the fronting of /uw/ and /ow/,[10] and stages [7] and [8] constitute a chain shift before /r/ in the back vowels, which is not part of the discussion of this paper.

 Bailey 1997 shows the same history for stages [2] and [6] as for [1]. However, [5] the fronting of /uw/ (and /u/) is reported as appearing variably in the earliest period, before 1875, and consistently from 1875 onward, and appears to have been temporally the earliest stage of the Southern Shift.

 Of all these features of the Southern Shift, the one that satisfies most clearly the minimax criterion outlined above is the monophthongization of /ay/. Map 1 shows that it is not found in any speaker north of the Southern line, and it is found for all speakers South of the line. Several points must be made to clarify the pattern here.

 (1) Map 1 does not report the frequency of monophthongization. However, the majority of our speakers show 100% or close to it; a lower percentage is found in the largest Texas cities, Dallas and Houston. Two of the northernmost speakers show a minimal amount of monophthongization, in Springfield, MO and Richmond, VA.

 (2) A small amount of monophthongization is found among speakers north of the line, in the South Midland and in Pennsylvania, but this is always before liquids (miles, Irish) or nasals (kind of). All the speakers south of the line, including those with very low percentages, show some monophthongization before oral obstruents.

 (3) The sub-area of the Coastal Southeast, to be discussed below, shows a lower degree of this feature. The Savannah speaker uses 37% monophthongs, and the Charleston speaker only before /l/ (in islander).

Again, we can observe that almost all of the elements of the Southern Shift enter into the solid configuration we find here. Table 2 sows the co-occurrence of a number of stages of the Southern Shift. No occurrences were found in any other region besides the South, the Coastal Southeast, and the South Midland.

 

TABLE 2
 CO-OCCURRENCE OF SOUTHERN FEATURES
 
Total number of speakers  The South 

25 

Coastal SE 

 2 

So. Midland 

 37 

Monophthongization of /ay/ before 

 voiced consonants and finally 

25 
Monophthongization of /ay/ before 

 voiceless consonants 

16 
Cartesian distance between /ey/ and 

 /ay/ < 500 

18  1* 
Inglides after /i/ and /e > 20% 
Reversal of positions of /ey/ and /e/ 
Reversal of positions of /iy/ and /i/ 
Back upglides with /oh/ > 0%  19 
Back upglides with /oh/ > 20%  15 
*Denver, Colorado

 

As noted above, monophthongization before voiced consonants and finally is the unmarked form throughout the South, and defines the region. The second line of Table 2 shows that only 16 of the 25 speakers use monophthongal forms before voiceless consonants. This is well known to be a socially marked feature: those who freely use monophthongal forms in why and ride may hesitate to do so in night and like. The geographic distribution of this feature corresponds to this, since the coastal South -- in the East and the Gulf States -- is where we find none of these forms before voiceless consonants. The older patterns of higher prestige Southern speech were concentrated in those coastal areas (McDavid 1964).

 Since the lowering of [ey] in Stage 2 depends upon the progress of this monophthongization, it stands to reason that the speakers who show the extreme form of it will be a subset of those who monophthongize. A direct measure of this lowering is the Cartesian distance between the nuclei of /ey/ (checked) and /ay/ (before voiced consonants and finally). Table 2 shows that 18 of the speakers in the South had values < 500, and only one speaker outside the South had this feature. The exceptional case is from Denver, Colorado, a city that shows a sizeable proportion of Southern-influenced speakers, just to the west of the South Midland. We further expect that the downward trajectory of /ey/ is dependent on the completeness of the monophthongization of /ay/. This is confirmed by the fact that all of the speakers who show a distance between /ey/ and /ay/ of less than 500 show more than 60% monophthongization of /ay/ before voiced consonants and finally.

 The occurrence of stage [4] of the Southern Shift, with the raising and fronting of /i/ and /e/, can be detected by examining the relative positions of /e/ and /ey/. Table 2 shows that nine speakers in the South showed this feature, as did three from the South Midland (from the western portion, in Kansas). These nine are concentrated in a coherent area indicated by the dashed line in Map 1 that includes most of the Appalachian area and the Piedmont, excluding the Coastal South. We can conclude that this area is the heartland of the Southern Shift, a coherent process that involves to one degree or another the entire area delineated in Map 1.

 A further implicational relationship concerns the full reversal of the relative positions of /i/ and /iy/, which is considerably less frequent and depends upon the prior descent of /ey/. Table 2 shows that there were only three such speakers located within the inner dotted ellipse of Map 1, along the Appalachian Mountains.

 In Table 2, we can see that a small area of the Southeast is distinct from the rest of the South, including the two cities of Charleston and Savannah. Though this area shows a low level of monophthongization of /ay/, none of the other features of the Southern Shift appear. This stems from the basic characteristic of the Coastal Southeast that separates it from the first division between North and South seen at the top of Figure 1, the laxing of the long high and mid nuclei. The distance between /ey/ and /ay/ is greater than 800 for the two speakers , as opposed to less than 500 for the surrounding territory. This is the feature used on Map 1 to define the Charleston-Savannah area as shown by the the two red crosses symbols.

 Another highly characteristic feature of the Southern region independent of the Southern Shift involves the nature of the distinction between /o/ and /oh/ in cot vs. caught, Don vs. dawn. While in other dialects this distinction depends upon vowel quality, for most Southern speakers the F1/F2 positions of the nuclei of these vowel classes are identical. /oh/ is distinguished from /o/ by a back upglide, so that it is phonologically /Ow/ opposed to /ow/ in height. Table 2 shows that this feature is strongly represented in the South, in 19 of the 25 speakers. If we raise the criterion to more than 10% back upglides, there are still 15 points represented. The back upglides are absent at the margins of the South: in west Texas, in Kentucky and Virginia, and in the Southeastern city of Charleston.[11]

 The Midland

 

We are now ready to follow the right hand branch at node B on Figure 1: the Midland dialects that lax the long high and mid nuclei but retain diphthongal /ay/. The first criterion differentiates the Midland from the North Central region; the second, from the South. We have already seen that the Midland region is sharply defined by its lack of participation in the Northern Cities Shift and the Southern Shift As the intimate relations of North Central and Inland North demonstrate, there is an inherent connection between the retention of tense nuclei and the Northern Cities Shift.. It follows that the Midland is distinct from both the North Central area in that the nuclei of the long high and mid vowels are not the conservative tense vowels characteristic of that region.[12]

 The Midland itself is divided into two sections: South and North. In the review of the South, we saw in Table 2 that there were many Southern features that were represented to a small extent in the South Midland; none appeared in the North Midland. The South Midland is not to be confused with the South Midland of Kurath and McDavid 1961, since it does not include the Appalachian region. Map 2 registers a roughly equal North/South division in the remaining Midland area. The criterion for South Midland membership is a simple one: the fronting of checked /ow/ as registered by an F2 of greater than 1350 Hz. Table 3 shows that there is no representation of this feature in the North Central region or in the Inland North, and only three points in the North Midland out of 36. The South Midland shows 21 out of 37 speakers with the fronting of checked /ow/, an internal consistency of 57%. Map 1 does not show these shaded triangles for the South, since the simple for diphthongization of /ay/ overrides this criterion, but all but two speakers are included. The exceptions are Louisville, Kentucky and Savannah, Georgia. If we combine the South Midland and the South, the external consistency of this criterion is 78%. The distribution of this feature in particular Midland cities indicates that Philadelphia is a member of the South Midland, while Pittsburgh and St. Louis can be assigned to the North Midland. .

 

Table 3
 Distribution of the defining feature of the Midland
 
NC  Inl 
No. 
No. 
Mid 
So 
Mid 
West  So.  St. L  Pitts  Phil  ENE  WNE  Tot 
N:  35  54  33  35  24  27 
F2 /owC/ < 1100  21  25  59 
Absence of marked features  20  41 

The North Midland

On Map 1, there is no single defining feature of the North Midland given. In fact, the most characteristic sign of North Midland membership on this map is the small black dot that indicates a speaker with none of the defining features given. Table 3 shows the numbers of speakers who are recorded on Map 1 in this way: 20 of the 33 North Midland points are included, a much higher percentage than for any other region. In terms of the analysis presented so far, the North Midland is not the most unmarked dialect. As we noted above, northern Iowa comes closest to the initial position of American English, but southern Iowa is perhaps closer to the default value.

The West

It is generally agreed that the diversity of American dialects declines steadily as one moves westward, and that the western part of the United States shows a diffusion of Northern, Midland and Southern characteristics as a result of the northward movement of Southern speakers through the Dakotas and Montana, and the steady flow of people from various regions to the far western states. Telsur data confirms this view in many ways. The conservative position of free /ow/ extends northward from the North Central states to Washington. Southern influence appears in the pronunciation of on with tense /oh/, and the use of inglides with the short vowels is widespread throughout the West, and particularly in Montana. But in many other respects, the West is aligned with the Midland.

 In Figure 1, this joint heritage of the West is shown by lines descending from both the Midland node and the pre-Northern node C. However, the end result of this mixture for urban dialects is not the diffuse or unmarked character that was predicted for the West. Instead, we find that in the phonological system, a fair degree of homogeneity is emerging, with specific features that differentiate the West from other dialect areas.

The most prominent feature of Western phonology is the merger of long and short open /o/. This is far from unique to the West. The Atlas records of Kurath and McDavid 1961 located the merger in Eastern New England and Western Pennsylvania. The first national map of any dialect feature was the product of work with long distance telephone operators in the 1960's to delineate the extent of is merger (Labov 1991:Figure 1.12). This map confirmed the Atlas records of merger in Eastern New England and Western Pennsylvania, with some evidence of expansion in both areas. It showed a solid merger in the West, with some variation in Los Angeles and the Bay area. The transition zone extended through Minnesota, Iowa, Nebraska, Oklahoma, the panhandle of Texas and southern Arizona and New Mexico. Since then, research has documented the progress of the merger in independent sites Eastern Pennsylvania (Herold 1990), Salt Lake City (Di Paolo 1988, Di Paolo and Faber 1990), Oklahoma (Bailey, Wikle, Tillery and Sand 1993) and Texas (Bailey, Wikle and Sand 1991).[13]

 Map 3 shows all Telsur speakers who pronounce cot and caught as the same in the phonetician's judgment. The same dialect boundaries of Map 1 are preserved, so that the distribution of this feature by region can be seen. It is immediately obvious that the merger is solid in the West: the only exception in the 24 speakers is one person in Denver. A second area of merger appears in a sub-section of the North Central region, in Northern Minnesota and Wisconsin: 7 speakers enclosed in a dotted line on Map 3. The 7 speakers in the Western Pennsylvania region around Pittsburgh all show the merger, with signs of a further expansion in Southern Ohio. The Eastern New England speakers show merger, but only one speaker in Western New England. Evidence for the expansion of the merger in Texas in Bailey, Wikle and Sand 1991 is supported by two Texas speakers in the Telsur area. Elsewhere in the South, the distinction is intact except for the region of Northern Appalachia: Lexington, Kentucky and Huntington, West Virginia. The complete Telsur records give further support to the finding that a new expansion of the merger is to be found in the Appalachian area.[14] Another center of merger appears in South Midland Kansas, which may be an expansion of the Western region. Though our single Oklahoma speaker does not show the merger in this pair, Bailey et al. 1991 show that expansion of the merger is quite general through Oklahoma, and Oklahoma, and the Kansas group may be a part of the same configuration.

At first glance, it seems that the merger is beginning to spread over all of the United States. But a closer examination shows that several large regions show almost total resistance to this sound change. The Inland North as a whole is the most prominent in this respect, a fact which we relate directly to the forward movement of short /o;/ in the Northern Cities shift. On the other hand, the Mid Atlantic States show no trace of merger, a fact which we relate to the upward shift of /oh/. But the Midland, and especially the South Midland, does not participate in either of these shifts which would increase the distance between /o/ and /oh/, and the result is shown as considerable instability.

 In all of the regions of merger, the West is the most extensive. Table 4 shows that the internal consistency of the West is 23/24 or 96%.[15] On the other hand, one can hardly say that the merger defines the West: the external consistency is 23 out of 64, or only 36%. We therefore consider a second variable, the fronting of free /uw/ in too, two, do, etc.[16] The second line of Table 4 shows the distribution of all speakers whose mean value for F2 of free /uw/ is greater than 1850 Hz. More than half of the speakers satisfy this criterion: 152 in all. It include speakers from almost every area. However, most of those who use front /uwF/ do not have a merger of /o/ and /oh/. The third line shows the result of combining the two criteria. Only 39 speakers are left: none from the North Central or Inland North, St. Louis, Philadelphia, or New England. The only sizeable group of speakers outside of the West is in the South Midland. For the South Midland, the internal consistency of this combination is 11/35, or 31%. For the West, it is 19/24, or 80%. The external consistency then rises from 34% to 49. This sharpens the picture considerably. It amounts to saying that this combination is characteristic of the west, but not uniquely or even predominantly so. This is the set of defining features represented by a star on Map 1.

 The picture of the West that emerges is that it has developed a characteristic but not unique phonology. It is closest to the South Midland as a dialect in which the merger of /o/ and /oh/ is firmly coupled with the fronting of /uw/. Though there are many speakers in the South Midland who show this pattern, it is still only a third of those we have interviewed, as opposed to four fifths of the Westerners.

 The combination of the /o~oh/ merger and the fronting of /uw/ is not to be taken lightly. For a considerable period, the only substantial theory concerning the conditions for chain shifting was that proposed by Martinet (1952) and supported by the work of Haudricourt and Juilland (1949): that the fronting of /u/ and /o/ was the response to overcrowding among the back vowels. More specifically, four degrees of height are common and stable among front vowels, but four degrees of height in the back vowels leads to either merger or the forward movement of the high back and or high mid vowels. It follows that the low back merger of /o/ and /oh/, which reduces the number of distinctions of height in the back to three, should be incompatible with the fronting of /uw/. In fact, Table 3 does show that for the great majority of American speakers, this is so. For most regions, the two tendencies are in complementary distribution. It is only in the West that there is a correlation between the fronting of /uw/ and the low back merger. Whatever the factors may be that lead to this combination, it is not accounted for by any current theory, and represents a distinct phonological configuration not general for any other American dialect.

 

Table 4
 Distribution of the defining feature of the West
 
NC  Inl 
 No. 
No. 
 Mid 
So 
 Mid 
West  So.  St. L  Pitts  Phil  ENE  WNE  Tot  Int%  Ext 

 % 

N:  35  54  33  35  24  27 
a) ot_oht  12  23  67  96  34 
b) F2/uwF/ > 1850  14  15  30  21  24  31  152  88  14 
a) and b)  11  19  39  79  49 
Returning to Figure 1, We see that the organization of American dialects is on the one hand dictated by their provenance, the descending, tree-like pattern. It is also the result of new features spreading geographically across traditional dialect boundaries. Prototypical of these is the fronting of /uw/, which now encompasses the South, the South Midland, the North Midland, and the West. Within this area we see the expansion of the fronting of /ow/, which governs the South and the South Midland, and in fact defines the South Midland as that region that shows the fronting of /ow/ without participating in the other aspects of the Southern Shift.

 

The Midland cities

It was stated above that the development of American dialects is controlled by two powerful and integrated sets of linguistic changes, the Northern Cities Shift and the Southern Shift, which drive the dialects of the Inland North and the South to ever more extreme forms in diametrically opposed directions. We recognize that the Midland has a very different and less organized character. Each of the Midland cities -- Philadelphia, Pittsburgh, Columbus, Cincinnati, Indianapolis, St. Louis, Kansas City -- has its own local character. Each urban dialect itself has developed local tendencies that are uniform throughout the city proper but shared only a limited degree by the hinterland. In Philadelphia we find the regional patterns of the fronting of /uw/, /ow/ and /aw/, the raising of /ahr/ and /ohr/ in a chain shift, and the centralization of /ay/ before voiceless consonants. But only within the city proper do we find the Philadelphia split of /æ/ into tense and lax categories, the near-merger of /e/ and /^/ before intervocalic /r/ in ferry and furry, and the reversal of the direction of the glide in /aw/ from [au] to [eO]. Similarly, the entire area of Western Pennsylvania, centered round Pittsburgh, shares the merger of /o/ and /oh/ the fronting of /uw/ and /ow/, and the vocalization of /l/. But only Pittsburgh itself shows the characteristic monophthongization of /aw/ in house, mountain, etc. Cincinnati shows a traditional split of /æ/ into tense and lax categories not dissimilar to that of New York City (Strassel and Boberg 1995). Here the pattern appears to be eroding in favor of a more general regional pattern of laxing.

 The city of St. Louis is located squarely in the South Midland region, but it has long been recognized as a center of Northern linguistic influence. On most Atlas maps, the St. Louis speakers show features that are held in common with the North, notable particularly in the long high and mid vowels, and there a corridor of northern influence that runs from northern Illinois to St. Louis (see Map 1). However, the specific configuration of St. Louis vowel system is local to the city in several respects. The most remarkable of these is a merger of /ahr/ and /ohr/ in card and cord, usually at the level of the mid vowel. This merger appears to be waning among younger speakers, and the vowel system seems to be shifting even more in the direction of the Inland North.

 The Phonological Atlas is only beginning to explore the local patterns of Midland Cities, and the notes on Figure 1 are only indicated as instances.. However, this local diversity may not be as exceptional as it seems at first glance. Given the vast regional patterns of the Northern Cities Shift and the Southern Shift, we tend to view the Midland heterogeneity as the marked case. But if we return to the great cities of the Eastern United States, we find this pattern of local diversity more than the sweeping regional developments. Thus Boston is the center of Eastern New England, but the fronting of the nucleus of vocalized /ahr/ and the inglide of short /o/ in /ohr are local to the city. Providence, as we have seen, has its own local pattern. Though New York City is coextensive with its hinterland, it is parallel to Boston and Philadelphia in developing its own local pattern, which does not spread far outside the city limits. The South, as we have seen, has a strong and uniform regional character, but we must also recognize that Richmond, Savannah and Charleston have their unique local urban pattern.

 The new and striking phenomena in American English are not the development of specific urban dialects, but the intensification of the great regional patterns. To the extent that Rochester, Buffalo and Chicago share the features of the Northern Cities Shift, the task of description is simplified, and the application to practical problems is facilitated. The same can be said for the opposing Southern Shift. It is even more relevant to note that we have not yet determined the nature of the forces that drive these sound changes, or explained why such sharp dialect boundaries persist over such long periods of time.

 

References
 

Ash, Sharon A.

Bailey, Charles-James N. 1968. Is there a "Midland" dialect?" ED 021 240.

 Bailey, Guy 1997. When did Southern American English begin? Ms.

 Bailey, Guy et al. 1991.. The Focus of Linguistic Innovation in Texas. English World-Wide 12(2): 195-214.

 Bailey, Guy, Tom Wikle, Jan Tillery and Lori Sand 1993. Some patterns of linguistic diffusion. Language Variation and Change 5:359-390.

 Boberg, Charles 1995. Evidence for the location of a North Midland dialect boundary: the pronunciation of on. Paper presented to the Midwest Modern Languages Association Annual Conference, St. Louis.

 Carver, Craig M.. 1987. American Regional Dialects: A Word Geography. Ann Arbor: U. of Michigan Press.

 Clarke, Sandra, Folrd Elms and Amani Youssef. 1995. The third dialect of English: Some Canadian evidence. Language Variation and Change 7:209-228.

 Di Paolo, Marianna 1988. Pronunciation and categorization in sound change. In K. Ferrara et al. (eds), Linguistic Change and Contact: NWAV XVI. Austin, TE: Dept of Linguistics, U. of Texas. Pp. 84-92.

 Di Paolo, Marianna and Alice Faber. 1990. Phonation Differences and the phonetic content of the tense-lax contrast in Utah English." Language Variation and Change 2:155-204..

 Eckert, Penelope. 1986. The roles of high school social structure in phonological change. Chicago Linguistic Society.

 Eckert, Penelope 1988. Adolescent social structure and the spread of linguistic change. Language in Society 17:183-208.

 Eckert, Penelope. 1989. The whole woman: Sex and gender differences in variation. Language Variation and Change 1:245-268.

 Eckert, Penelope. 1991. Social polarization and the choice of linguistic variants. In P. Eckert (ed.), New Ways of Analyzing Sound Change. New York: Academic Press.

 Fasold, Ralph. 1969. A sociolinguistic study of the pronunciation of three vowels in Detroit speech. Mimeographed.

 Haudricourt, A. G. and A. G. Juilland. 1949. Essai pour une histoire structurale du phonétisme français. Paris: C. Klincksieck.

 Herold, Ruth. 1990. Mechanisms of merger: The implementation and distribution of the low back merger in Eastern Pennsylvania. U. of Pennsylvania dissertation.

 Hindle, Donald. 1978. Approaches to vowel normalization in the study of natural speech. In D. Sankoff (ed.), Linguistic Variation: Models and Methods. New York: Academic Press.Pp. 161-172.

 Kurath, Hans and Raven I. McDavid, Jr. 1961. The Pronunciation of English in the Atlantic States. Ann Arbor: U. of Michigan Press.

 Kurath, Hans. 1949. Word Geography of the Eastern United States. Ann Arbor, MI: U. of Michigan Press.

 Labov, William 1994. Principles of Linguistic Change. Volume 1: Internal factors. Oxford: Blackwell Publishers.

 Labov, William 1996. The organization of dialect diversity. Paper given at ICSLP4, Philadelphia, November 1996. Phonological Atlas of North America, http:/www.ling.upenn.edu/phono_atlas/home.html

 Labov, William, Malcah Yaeger & Richard Steiner. 1972. A Quantitative Study of Sound Change in Progress. Philadelphia: U. S. Regional Survey.

 Labov, William. 1966. The Social Stratification of English in New York City. Washington D.C.: Center for Applied Linguistics.

 Labov, William. 1989. The limitations of context. CLS 25, Part 2, pp. 171-200.

 Labov, William. 1991. The three dialects of English. In P. Eckert (ed.), New Ways of Analyzing Sound Change. New York: Academic Press. Pp. 1-44.

 Martinet, Andre 1952. Function, Structure, and Sound Change Word, April 1952, Vol. 9, 1.

 McDavid, Raven I. Jr. 19. Postvocalic /r/ in South Carolina: A Social Analysis. In D. Hymes (ed.), Language in Culture and Society. New York: Harper & Row. Pp. 469-482.

 Nearey, Terence. 1977. Phonetic feature system for vowels. University of Connecticut dissertation.

 Shuy, Roger. 1962. The Northern-Midland dialect boundary in Illinois. Publication of the American Dialect Society 38:1-79.

 Strassel, Stephanie and Charles Boberg 1995. The reversal of a sound change in Cincinnati. Paper given at NWAVE 24, Philadelphia.

 

[1] The research reported here has been sponsored by the National Science Foundation under Contract FBR 92-22458, the National Endowment for the Humanities under RT 21599-95, and Nortel Technology.

 2 The PLOTNIK algorithm for calculating means excludes words with /l/ codas, /w/ and /y/ onsets, and obstruient-liquid onsets. Vowels before tautosyllabic /r/ are coded as separate vowel classes.

 [3] One divergence within this initial position concerns the low back vowel /oh/. In most of the South this vowel is differentiated from /o/ by a back upglide and is phonologically /Ow/.

 [4] In the North Central States, the Pennsylvania German area, the Coastal Southeast, and in several other dialects influenced by a Scandinavian, Celtic or Caribbean substratum, these vowels may be monophthongal or even ingliding.

 [5] The West does not disregard this division entirely. In future maps it will be shown that a clear North/South division persists in the Western states where the laxing of the high and mid long vowels is much less common in the northern third.

 [6] Note that the "conservative" position of free /uw/ is actually considerably more advanced than that for the /ow/ allophones. The fronting of free /uw/ is a process that has begun to affect almost all American dialects in recent decades, to some degree or other.

 [7] For the purposes of this tabulation, Charleston and Savannah are included in the South, and Providence is included in "Western New England", although they form distinct units geographically and phonologically.

 [8] Several additional points on Map 1 actually satisfy the Inland North criterion, but show a different symbol to designate specific local features. St. Louis, which shows strong Northern influence in many respects, includes three speakers with this approximation of the front-back positions of /e/ and /o/.

[9] Bailey et al. 1991 refer to this phenomenon as "glide shortening" throughout, and preserve this term in other papers, although the term "monophthongal /ai/" is reintroduced in Bailey 1997. Their view is based upon an earlier study that showed a number of truncated glides where one would expect complete monophthongization. Our own acoustic analyses of Southern English show that such shortened glides are the exceptional case before voiced obstruents and finally. For each occurrence of monophthongal /ay/, the analyst examined the spectrogram and LPC tracing for any indication of an upward shift in F2 or downward shift in F1 before the transition, and then listened to various sections of the resonant portion of the vowel. If no trace of a glide was detected, the token was labeled "{m}"; if there was indication of a truncated glide, "{s}" The overwhelming majority of such tokens showed complete monophthongization.

 [10] The fronting of the nucleus of /aw/ might well be included in this pattern, as a third element in the parallel movement of /uw/ and /ow/. It is another feature that clusters tightly along the North/North Midland line: Inland North dialects generally show /ay/ with a nucleus fronter than the nucleus of /aw/.

 [11] Although the Southern Shift is a coherent pattern that appears to be developing and diverging from other dialects, a contrary movement can be detected in certain large cities strongly affected by the large inmigration of Northerners. Thomas (to appear) finds such a contrast between the major metropolises of Texas and the rest of the state. Our data from Dallas confirms Thomas' finding. The frequency of all features of the Southern Shift in Dallas and Houston is diminished among young native speakers in Dallas and Houston, as compared to smaller cities like Lubbock and Odessa, and this appears to be true in the cities of the Research Triangle of North Carolina, Winston-Salem, Durham and Raleigh

 [12] Note that the fronting of /uw/ is not necessarily associated with the laxing of the /u/ nucleus, since it moves the nucleus frontward across the upper periphery of the phonological space involved. Thus as we will see, the fronting of /uw/ is a tendency common to almost all dialects of American English, including those that retain tense /ey/, /iy/ and /ow/ nuclei and those that do not.

 [13] The merger of /o/ and /oh/ follows Herzog's principle that mergers must expand geographically at the expense of distinctions (Labov 1994:, Chapter 10). However, the first Telsur results given in the maps of Labov 1996 do not show the expected Eastward shift of the transition zone. It appears that the Northern Cities Shift provides a substantial structural impediment to the expansion of the merger.

 [14] Feagin 19?? reported that some Anniston upper middle class speakers were showing a merger of /o/ and /oh/ as they dropped the traditional back upglide.

 [15] If we take a much stricter criterion, and ask how many speakers produced /o/ and /oh/ in all environments as the same in formal elicitation, and judged them as the same, the result is 18 out of 24, or 75%.

 [16] In actual fact, the more effective definition of this class is /uw/ after coronal onsets. In most of the vocabulary dealt with here, this is equivalent to "free /uw/" or uwF, and this term is preserved.