<?xml version="1.0" standalone="yes"?>
<Paper uid="C69-5201">
  <Title>CC~fl~TER-PRODUCED REPRES~EATION OF DIALECTAL VARIATION: INITIAL FRICATIVES IN SOUTHERN BRITISH ~GLISH</Title>
  <Section position="1" start_page="0" end_page="6" type="metho">
    <SectionTitle>
CC~fl~TER-PRODUCED REPRES~EATION OF DIALECTAL VARIATION:
INITIAL FRICATIVES IN SOUTHERN BRITISH ~GLISH
</SectionTitle>
    <Paragraph position="0"> W. Nelson Francis, Jan Svartvik, and Gerald M. Rubln Department of Linguistics, Brown University It has become apparent to some dialectolo6ists that di~lectology, particularly in its interpretive phase, is a branch of linguistics particularly adapted to the use of computers. The dialectol0gist typically deals with large bodies of data, usually in the form of single words and short phrases, and he is interested in sorting and comparing individual items on many bases: phonological, morphological, lexlcal, and geographical. The major obstacle that has prevented widespread use of computers in dialect study is the fact that the data for most of the great dialect surveys have been collec-I ted, recorded, and in most cases edited prior to the computer age. Thus the problem of preparing large bodies of data, much of it in narrow phonetic transcription, for computer use has been formidable.</Paragraph>
    <Paragraph position="1"> One of the aims of the present paper is to show that results can be obtained relatively easily by computerized sorting and mapping that would take endless hours by traditional methods, and hopefully to encourage others to invest time and money in preparing data for the computer rather than in time-cc~suming hand sorting and map-making.</Paragraph>
    <Paragraph position="2"> Accordingly we sought a proble~ I that would be canplex enough to reveal the advantages of computerized dialectolo~y while at the same  time involving a body of data small enough %0 be quickly prepared.</Paragraph>
    <Paragraph position="3"> Since two of the three,authors are specialists in English (the third is a computer specialist), we ~tural/y ~urned to the published volumes of the 5'u~e~ d of ~dZ'~h D~a/~c~, 2 which embody carefully controlled data, collected with professional skill, and presented in convenient tabular form in meticulously edited and printed volumes.</Paragraph>
    <Paragraph position="4"> And since une of the two areas coveredby the volumes in print at the time the study was ~adertaken (May 1969) was the south of England, the problem of the voicing of initial fricatives in the southwest naturally suggested itself. This lyrohlem had the further advs~tage, for our purposes, of dea//ng with cm~sonants (s inkier than vowels in most varieties of English) in /D/tial position, hence easily sorted and exsmined. The selection of this proble~ has proved to be a happy one.</Paragraph>
    <Paragraph position="5"> The area covered by Volume ~ of SED comprises the ten southernmost counties of England, which, with their key numbers in the S~w;ey, are 31 Somersetshire, 32 Wiltshire, 33 Berkshire, 3~ Surrey, 35 Kent, 36 Cornwall, 37 Devonshire, 38 Dorsetshire, 39 Hampshire, and h0 Sussex. A :~ &amp;quot;~i of 75 localities in this area were reported os by the Su~ey; in what follows these will be identified by a four-digit r number, the first two digits indicating the county and the remaining two the locality. Thus 3906 stands for Burley, the sixth locality listed in hampshire, accordi~to the numbered list on p. 31 of the Introduction to the 8ul-oey. The data selected for examination  included all those words beginning with graphic f-, s-, or thfollowed by a vowel or voiced consonant which were starred in the ~ED questionnaire. 3 To this list we later added a few non-starred words which showed universal distribution and were otherwise of interest. The final list contained 68 words, of which 27 are f-words, 22 s-words, 16 Sh-words and 3 ah-words (i.e. words beginning with /~/ in standard English). We took only the first recorded form from each locality; this is presumably a citation form, produced by 8n informant in response to a question, and recorded in narrow IPA transcription. The 59 cases where no response was given were coded XXX in our computer code.</Paragraph>
    <Paragraph position="6"> The corpus thus comprised 68 x 75 or 5100 items, including the 59 blanks. Our c~,puter expert then produced 68 decks of punch-cards, one for each word~ each deck containing 75 cards, one for each locality, These were numbered at the left for locality and on the right for the reference number of the item in the SED questionnaire. ~ A coding system was devised which preserved all significant features of the phonetic transcription while passing over apparently irrelevant fine points (see Appendix A), and the words were transcribed in this code directly onto the cards for the guidance of the key-puncher, who then punched the coded words in a fixed place on the cards.</Paragraph>
    <Paragraph position="7"> Subsequently the standard spelling was inserted by the computer to the left of the coded phonetic spelling. This whole process took about a dozen hours of the investigators' time (not counting the relatively simple progras~ing involved) and about the same amount of the key-puncher's time. The result was a body of data consisting of 5100 entries of the following sort: 3101 FINGER FIgG)R. 6 7 7 This is to be interpreted as indicating that at locality 3101 (Weston in Somerset) the word f~nger, which appears as item VI.7.7 of the SED questionnaire, is pronounced \[f,ngar-\] (or perhaps more accurately /ftngar-/ in the quasi-phonemic transcription used).</Paragraph>
    <Paragraph position="8"> The nature of the problem with which we are dealing ms~ be most simply introduced by an excerpt frc~ the full treatment given to the voicing of initial fricatives in Middle English by Horn and Lehnert (195~, Vol. If, ~37): In gewissen Mnndarten sind in alter Zeit im Wortanlaut die starken und stimmlosen Reibelaute f, s und ~ vor Vokalen und schwachen, stimmhaften Konsonanten schwach und stimmhaft gew0rden: f- * v-, s- &gt; z-, ~- * 6- . . . Die neuenglische Hochsprache hat einige ~6rter aus den Mundarten aufgen~n...</Paragraph>
    <Paragraph position="9"> Die starken und stimmlosen Reibelaute im Wortanlaut sind in Kent und im sUdSstlichen Mittelland schwach und stimmhaft geworden. In me. Handsehriften wird v f'dr anlautendes ~ vor vor Vok~len geschrieben, z f~r enlautendes e vor Vokalen, . . .</Paragraph>
    <Paragraph position="10"> Wit d~rfen annehmen, da~ aueh stimmloses th- stimmhaft geworden ist, da heutige Mundarten f'dr ~- ebenso wie f~r f- ,rod sI null stimmhafte Laute haben...</Paragraph>
    <Paragraph position="11"> Aus der Tatsache, da~ franzSsische Lehnw~rter im Me. ~- mud s- beibehalten . . . geht hervor, da~ die stimmlosen Reibalaute in den englischen WSrtern schon vor der Aufnahme der franzSsischen LehnwSrter stimmhaft geworden waren...</Paragraph>
    <Paragraph position="12">  Der Lautwandel hat sich im Laufe der Zeit von Kent aus h'oer die s~dliehen und angrenzenden ~stlichen Grafschaften ausgedehnt, ~uld zwar hat er heute in dlesen Gebieten anlautendes ebenso ergriffen wie fund s. Vereinzelt ist auch anlautendes ~, zu ~ geworden... In den heutigen Mundarten yon Kent, Surrey und Sussex sind die anlautenden stinunhaften Laute unter dem Einfluss des Hochenglischen durch stimmlose ersetzt worden, w~d das ~stliche Herefordshire, Teile von Gloucestershire, das westliche Berkshire, und besonders Wiltshire, Dorsetshire, Scaersetshire told Devonshire stinunhafte Reibelaute im Anlaut aufweisen.</Paragraph>
    <Paragraph position="13"> Horn-Lehnert go on to point out that by analogy and under the influence of Standard English, initial f- and s- in French loanwords have become voiced. This is borne out in our corpus by the words ~, ~Grmer, and ~ers, which have initial v- in more than half the localities included. As our subsequent discussion and maps will show, Horn-Lehnert should have included in the initial voiced fricative area the western half of Hampshire and all but the southwest tip of Cornwall. Since the SED records for Herefordshire and Gloucestershire have not yet been published, we have not been able to include these counties in our survey.</Paragraph>
    <Paragraph position="14"> The traditional assumptions, then, are that ~-, s-jand ~-.</Paragraph>
    <Paragraph position="15"> bee~ne voiced (except before voiceless consonants) in initial position at an early date--certainly before the 13th century--in the southeast; that this affected all native words ; that this change subsequently spread into the old West Saxon area of the southwest ; that  after this spread voiceless initial fricatives were restored in the southeast; and that analogy, dialect borrowing, and the influence of standard English worked variously to blur the excepticmless character of the sound change, to produce voicing of initial 2- and ain French loanwords and of initial ~- in native words like 8hiZlir~ as well as French words like sure, and othe~se to create a mixed situation in the whole southern area. 5 Our project was to see what light the records of the SED can throw on this situation by exploiting them in some of the many ways made possible by computer technology. null The first step was to sort the data in as many ways as we felt would be productive. Accordingly our computer expert produced four lists of the 5100 items, sorted as follows : List I: sorted first by keyword (the standard English graphic word identifying the item) alphabetically; then by locality.</Paragraph>
    <Paragraph position="16"> This list presents the data in the same kind of order in which it appears in the SED records and allows easy inspection of all versions of each word in one list.</Paragraph>
    <Paragraph position="17"> List II: sorted first by locality and than by keyword alphabetically. 'Fais list brings together in one place all the records from each locality and thus permits comparison of the 8mount of voicing recorded fr~n various localities.</Paragraph>
    <Paragraph position="18"> List Ill: sorted first by locality and then by citation (the recorded local form) alphanumerically. This list even more  graphically reveals the -mount of initial voicing; it also makes it easy to surmise from inspection whether or not incidence of initial voicing might be influenced by the following vowel or cG~Is onsIlt.</Paragraph>
    <Paragraph position="19"> List IV: sorted first on the second and following characters of the citation form, in linguistically significant order (i.e. by vowels and consonants in articulatory order), then by the first character. This list greatly facilitates looking into the question of the possible influence of following sounds on initial voicing.</Paragraph>
    <Paragraph position="20"> These lists, though interesting in themselves, were cmm~idered primarily as intermediate diagnosti~ procedures, to be used to guide us in future ~ortlng, counting, and eventually mapping. Even the most cursory inspeetiou of them revealed what we had suspected frc~ our first look at the data: that there is tremendous variatio~ within the relevant area both from word to word and from locality to locali ty. It ~ertainly seemed, at least so far as this feature in this region is concerned, that the maxim attributed to Gilli~ron, &amp;quot;Every word has its own history,&amp;quot; is true.</Paragraph>
    <Paragraph position="21"> Accordingly we asked our c~nputer expert for various more sophisticated sortings and counts, and for two kinds of maps, to be produced by the CalCump plotter (see Appendix B). These were the following: i. A table for each of the four sets of words, listing all words  in the set and counting the number of each occnrrf~ initial consonant, the words to be ranked in descending order of the n~ber of initial voiced consonants recorded. ~ese are reproduced here as Tables i - 4.</Paragraph>
  </Section>
  <Section position="2" start_page="6" end_page="31" type="metho">
    <SectionTitle>
2. A list of all possible ~vowels, dipht~s, and second conson-
</SectionTitle>
    <Paragraph position="0"> ants in each set of words, counted and tabulated in terms of each possible initial consonant. A portion of this list is reproduced herewith as Table 5.</Paragraph>
    <Paragraph position="1"> 3. An individual map for each of the 68 words, showing the initial consonant recorded for each locality. Eight of these are reproduced here as Maps 6-13.</Paragraph>
    <Paragraph position="2"> 4. Four proportional maps, one for each set of words, indicating at each locality the voiceless:voiced ratios; thus for the ~h-words the legend 3/13 occurring at B905 indicates that at that locality only three of the 16 d-words begin with voiceless consauants. Map I shows the proportianal map for the e-words as it came from the plotter, while maps derived frc~ these proportional maps to reveal the varyiug distributions more clearly are included as Maps 2-5.</Paragraph>
    <Paragraph position="3"> Tables i - ~ s~pport our suspicion that each word has its own unique distribution with regard to the initial c~nsonant. Thus Table 1 uhows that ~ the f-words the proportion of voiceless to voiced ranges fr~ 20:5~ in F~.LT~ to 52:22 in FOAL. Even in those cases vhere the proportions are the same, recourse to List I ~eveals ~hat  Vcls. Vcd. Total Total Missing 1, FELLIES 1.9.9 1'9 5 C/ 2. FURROW 2.3.1 53 C/ C/ 3. FOOT 6.10.1 h6 C/ 1 h. FRIDAY-1 7.4.h h6 C/ C/ 5. FERN h.10.13 h5 C/ 6. FINGER 6.7.7 h5 C/ 7. FIRE 5.3.1 h5 ~ &amp;quot; 8. FRIDAY-2 7.4.7 4h I~ (~ 9. FIVE 7.5.6 h2 ~ i0. FLEAS h.8.h b2 @ ii. FROGS 4.9.6 h2 ~ @ 12. FIND 9.3.2 hl ~ 13. FOX h.5.11 )41 C/ C/ 114. FURTHER 9.2.1 b,l C/ C/ 15. FLIES h.8.5 .hO ~6 C/ 16. FLOUR 5.6.1 )40 C/ @ 17. FIRST 7.2.1 39 C/ 18. FLOOR 5.2.7 38 ~ i 19. FARMER 8.h.7 38 C/ C/ 20. FLITCH 3.12.3 38 C/ C/ 21. FLOWERS 8.5.13 36 C/ C/ 22. FORTy 7.1. lh 35 @ C/ 23. FATHER 8.1.i. 31 ~ 24. FIGHT 3.13.6 31 C/ 25. FORD h.i. 3 22 ~ i 26. FOAL 3. h. i 22 ,16 9 I  i. SATUROAX 7.4.5 46 C/ 29 ~ 29 46 C/ 2. S~E 6.3.2 46 ~ 29 C/ 29 46 C/ 3. six 7.1.5 42 l 32 ~ 32 43 C/ 4. SOW-N 3.8.6 43 @ 31 @ 31 43 1 5. SUCK 3.7.1 42 ~ 32 ~ 32 42 1 6. SEV~ 7.1.6 41 ~ 34 @ 34 41 @ 7. SADDLE 1.5.6 40 ~ 33 ~ 33 40 2 8. SOUTH 7.6.25 40 ~ 35 @ 35 40 9. SIGHT 8.2.9 39 ~ 36 ~ 36 39 i0~ SILVER 7.7.7 39 @ 35 ~ 35 39 1 ii. SEW 5.10.3 37 ~ 38 ~ 38 37 12. SOOT 5.4.6 37 ~ 38 ~ 38 37 13- SWEAT 6.13.5 33 1 41 ~ 41 34 14. SECOND 7.2.3 31 ~ 44 @ 44 31 @ 15. SUET 5.7.6 30 ~ 44 1 45 30 16. SWEAR 8.8.9 23 @ 52 ~ 52 23 g 17. SEXTON 8.5.h 22 @ 53 @ 53 22 @ 18. SNOW 7.6.13 15 1 59 @ 59 16 19. SMOKVS 5. i. 4 14 1 60 @ 60 15 @ 20. SNOUT 3.9.1 4 ~ 70 ~ 70 4 i 21. SLEDGE 1.9.1 3 @ 68 ~ 68 3 4 22. STITCH 5.10.4 i @ 74 @ 74 i @  i. THATCH-N 2.7.6 31 8 6 29 C/ 29 h5 1 2. THATCH-V 2.7.5 32 7 6 29 C/ 29 h5 i 3. THUMB 6.7.6 45 C/ C/ 30 C/ 30 h5 C/ h. THISTLE * 2.2.2 19 22 3 28 2 30 4h C/ 5. THIRSTY 6.13.10 h3 @ C/ 31 C/ 31 43 1 6. THIRTEEN 7.i.i ! 43 C/ C/ 32 C/ 32 43 C/ 7. THRESH 2.8.1 i 42 C/ 31 C/ 31 43 i 8. THURSDAY 7.4.3 43 ~ ~ 32 ~ 32 43 9. THREE 7.1.3 ~ 42 ~ 33 ~ 33 h2 i0. THIMBLE 5.10.9 4i ~ ~ 34 ~ 34 4i Ii. THIRD 7.2.h 4i C/ @ 3h ~ 3h hi 12. THIRTY 7.1.13 4I ~ ~ 31, @ 34 4i 13. THOUSAND 7.1.16 40 C/ C/ 35 0 35 40 lb. THREAD 7.10.2 ~ 39 ~ 36 ~ 36 39 15. THUNDER 7.6.21 39 ~ C/ 36 C/ 36 39 16. THIGH 6.9.3 33 ~ ~ ~ ~ Ill 33 i Word i. SUGAR 2. SHILLING 3. SURE</Paragraph>
    <Paragraph position="5"> the geographical distribution is not identical. Thus items 5 - 7 on the table all have a voiceless-voiced ratio of 30:45, but the voiceless localities, at least in the area to the west of the major isogloss to be described below, are not the same. FERN (see Map 13) is voiceless at 3102, 3604, 3902, and 3907 and voiced at 3405 and h003~ FINGER is voiceless at 3405, 3902~ 3907, and 4003 and voiced at 3102 and 3604; FIRE is voiceless at 3102, 3405, 3604, and 4003, and voiced at 3902 and 3907. In tabular form:  The same kind of discrepancy is shown by items 9- ii, with 33:42 ratio, and items 12 - 14, with 34:41.</Paragraph>
    <Paragraph position="6"> Table i also shows that neither etymology nor following consonant seems to affect the distribution markedly. The three French words, FLOUR, FARMER, and FLOWERS, appear in the middle of the list, with ratios of 35:40, 37:38, and 39:36 respectively. The words with initial ~- are also in the middle, ranging from FLEAS 33:42 to FLOWERS 39: 36.</Paragraph>
    <Paragraph position="7"> The three last words on the list, with ratios 41:23, 52:22, and 40:22, all show some peculiarity in the recordings indicating that for many of the informants they were unfamiliar, learned, or bookish words, hence more likely to have standard English pronunciation. Thus</Paragraph>
    <Paragraph position="9"> FORD, the expected response to the question &amp;quot;Sometimes there is no bridge (over a rivulet). What do you call that shallow place where you can walk across?&amp;quot; (SED IV.1.3), was not known, not found, or not recorded in seven localities, replaced by a local term (splash, sluice) in four others, and given as a &amp;quot;suggested word&amp;quot; (i.e. one pronounced first by the field worker) in nine others. Though FOAL is recorded from all but one locality in answer to III.~.l, it is revealed by the answers to another question (III.4.6) that in many localities the preferred word is colt. The records also show that FORKS (the agricultural kind), not recorded from 13 localities, yields to ~ other preferred local words (picks, prongs, spuds) in many others. Likewise it has a mixed etymology, being derivable from Latin furca either through OE forca or Anglo-Norman fourque.</Paragraph>
    <Paragraph position="10"> At the top of the list two words, FELLIES and FURROW, show unusual distribution in that voiced forms extend well into the usually voiceless areas of Berkshire, Sussex, and the tip of Cornwall (see Maps 6 and 7). FELLIES also shows six instances of the substitution of the dental fricative for the labial, which otherwise occurs only in FRIDAY, and which is reversed in six occurrences of /v/ in THATCH.</Paragraph>
    <Paragraph position="11"> If these five words are set aside, the range of voiceless-voiced</Paragraph>
    <Paragraph position="13"> ratios, from 28:~7 to 43:3~ much more closely resembles that of the th-words (Table 3), which range from 29:45 to 41:33.</Paragraph>
    <Paragraph position="14"> Observations in similar detail can also be made on the basis of  ii marked drop-off in voiced forms in the last five words of Table 2 implies that, in contrast to initial ~-, the voicing of initial 8is much reduced by a following nasal or /i/ (this will be discussed further below). Voicing is not to be expected before a voiceless consonant; we included STITCH because it shows one freakish occurrence of initial \[z\] at 3901. In this set the French words~ SECOND, SUE~f, and SEXTON, do appear well down the list, with ratios of ~4:31, ~5:30, and 53:22 respectively. If the words of French origin and those with a consonant after the initial 8- are set aside, the remaining range of ratios, from 29:46 to 38:37, comes much closer to that of the other sets.</Paragraph>
    <Paragraph position="15"> The th-words in Table 3 are, of course, all native words, since neither Norman nor Central French has had initial dental fricatives at any time when borrowing into English could occur. Likewise the only consonant that occurs at all frequently after initial /8/ or /~/ is /r/. Hence the range of ratios of the whole set is narrower than those of the f- and 8-words. Noteworthy is the fact that throughout the voiced area, the initial cluster /@r/ appears as /dr/, presumably as a result of a later change of /~/ to /d/ in this environment.7 (See Map 8 for THRESH). This same change has occurred in THISTLE in western Somerset, Devon, and Cornwall (see Map 9), and in THATCH (Map i0) in south Devon and eastern Cornwall. These two words also show initial /v/ sporadically in Somerset and Dorset, presumably a substitution the opposite of that occurring in FELLIES in five localities.</Paragraph>
    <Paragraph position="17"> One locality, 3805 (Kingston, Dorset), even has /v/ in THISTLE and I~/ in FELLIES. Why only two of the 16 %h-words should show this kind of substitutic~ remains unexplained. ~here is certainly nothing in their etymology to account for it.</Paragraph>
    <Paragraph position="18"> Table ~ lists the only three sh-words that show more than a few scattered instances of initial /~/&amp;quot; The other ~tarred words of this sort, with the number of occurrences of initial voicing for each, are 8heuf ~, s~ep i, she~f 3j 8hoe 2, and sh~;eZ 3. Even the three words listed have a ratio characteristic of the low end of the other lists: 52:22 for SUGAR (a French loanword), 57:18 for the native word &amp;quot;SHILLING, and 62:13 for another French word, SURE. Presumably the two French words had sc~ellhing like initial \[s J-\] at the time of borrowing, while 8~ILL~N~ and the other native words had /s~-/ until later OE. Instances of voicing in these words is to be attributed wholly to analogy, prest~ably with the s-words. As might be expected, the heaviest concentration of voiced forms is in the &amp;quot;hardcore&amp;quot; voicing areas in Devonshire and western Hampshire (see Map 5), but there are scattered instances in Sc~ereet, Dorset, and Wilts.</Paragraph>
    <Paragraph position="19"> (see Map ii for SHILLING).</Paragraph>
    <Paragraph position="20"> Just as Tables i - 4 reveal the wide variety in voiceless-voiced ratio for individual words, so Maps 2 - 5 show the same variety for individual localities. These m~ps have been simplified from the original cce~uter-produced proportional maps described above, which were in the form shown in Map 1. Taken together, they de~nstrate  several interesting points. All four of them show a strongly marked isogloss on the eastern boundary of the voicing area. This line starts o~ the coast east of Portsmouth, trends in a northwesterly directic~ across the middle of Hampshire, follows the Wiltshire Berkshire line for a short distance, and then turns vest across northern Wiltshire toward the Cotswolds. The plotting of its subsequent course must await the publication of the records fr~ Gloucestershire and Hereford. It marks a sharp division quite different frem the grading noticeable within the voicing area. At one point about ten miles north of Winchester it passes between two localities (3901 and 3902) showing respectively total voicing and total voicelessness in the th-words. A little to the north of this it separates two localities--Burbage, Wilts., and Inkpen, Berks.--which are only 8 miles apart and yet show voiceless voiced ratios of @:16 and 15:1 respectively in the th-words. It is equally well marked for the other sets, though in the case of the f-words the ~explicable eastern extension of voicing in two words--FELLIES AND FURR0W--creates the appearance of a transitional area (Map 2).</Paragraph>
    <Paragraph position="21"> The maps also show that the voicing did not extend to the south-west tip of Cornwall except in a few words. Presumably the English brought into this formerly Celtic area was more strongly influenced by standard English. The extreme case is represented by 3607, which has voicing in only two of the 68 words: FELLIES and FURROW, which, as we have seen above also extend beyond the voicing area on the east.</Paragraph>
    <Paragraph position="23"> Maps 2 - 5 also reveal rather similar patterns of voicing in different parts of the region. Just west of the main eastern isogloss is an area where voicing is virtually total for all sets of words. The area differs in size from one set to another, being larger for the f- and th-words than for the s- and sh-words, but its heart is western Hampshire, Dorset, and part of southern Wiltshire.</Paragraph>
    <Paragraph position="24"> It is separated from the other area of almost total voicing in Devonshire by a mixed area in Somerset and Wiltshire where the proportion of voicing among the words examined ranges from 25% to 75%.</Paragraph>
    <Paragraph position="25"> Finally there are fringe areas in the east and in southern Cornwall where ~ly a few of the words~those high on the lists in Tables 1 - h--show voicing. In the light of the traditional historical view of a feature spreading west from Kent, as ex~pressed in the quotation from Horn-Lehnert above, this distribution is a bit puzzling. Full exploration of its implications must be postponed to a later study.</Paragraph>
    <Paragraph position="26"> One possible explanation of the variation in the incidence of voicing from one word to another is to be sought in the influence of the sounds immediately following the initial fricative. Chart I displm~vs some of the characteristics of the four sets with regard to to following phonemes. The sh-words have far less initial voicing than the other sets. No more thou 22 localities have initial voicing for any single word in this set. On the other hand, the s-words demomStrate the widest range: frc~ voicing in 46 localities to a long tail of words with no initial voicing anywhere. The most stable is</Paragraph>
    <Paragraph position="28"> the th- set, which has between 45 and 33 localities with voicing.</Paragraph>
    <Paragraph position="29"> This means that for all words but one in this set, over half the localities have a voiced initial fricative. This is the highest proportion for any set.</Paragraph>
    <Paragraph position="30"> The sound following the initial fricative has been indicated on Chart I in terms of three categories: + vowel (unmarked) + voiced consonant (marked with square) + voiceless consonant (marked with dot) There is a striking correlation between voiceless initial fricative and a following voiceless consonant in the s-words (no other set has a voiceless consonant in second position). With the exception of spring (72:3) and STITCH (74:1), the association is absolute for Ik/ school, scratch, sky, squirrel (s) + /p/ spade, speak, spokes It~ stars, steal, stile, stool, straw  In the case of the other sets, however, there is no clear connection:  Vowels present a more complex situation. Table 6 shows that for f-, s-, and th-words, high and low front vowels /y, i; ~/ and low central vowels /a, a/ associate with initial voicing. For the three sets taken together the proportion of voiced initial fricatives occurring with these vowels following is between 60 and 63 per cent.</Paragraph>
  </Section>
  <Section position="3" start_page="31" end_page="32" type="metho">
    <SectionTitle>
TABLE 6
PROPORTION OF VOICELESS TO VOICED INITIAL FRICATIVES
IN RELATION TO THE FOLLOWING VOWEL IN F-, S-, TH-WORDS
</SectionTitle>
    <Paragraph position="0"> There is negative association between voicing and mid front /e, e, cPS/ as well as low back vowels /D, 3/. A study of t~e values for the individual sets reveals that initial voicing is particularly associated with high and low front vowels in f-words (72% and 78% respectively) and with low central vowels in 8- and th-words (77% and 83%).</Paragraph>
    <Paragraph position="1">  In f- and th-words low back vowels have a particularly low sssociati(m with initial voicing (20% and 30%, respectively). Vowel length appears to have no appreciable correlation with the voicing or unvoicing of the initial fricative.</Paragraph>
    <Paragraph position="2"> Rounded high front vowels /Y, y/ showed a more marked association with initial voicing than unrounded high front vowels /I, i/. The ratios for all f- and s-words with rounded front vowels are: FOOT @:16; SOOT 1:15, SUCK @:7, SUET 1:15. If we make a table for all the second vowels in SOOT (Map 12), we get:  There is, of course, a possibility here of pseudo-correlation, caused for instance by the possibility that those localities which have the fronted /Y/ for standard English /U/ in words like SOOT happen to be located in the area of strongest voicing. Investigation of this possibility must await further study.</Paragraph>
    <Paragraph position="3"> The tendency for high front vowels generally to co-occur with voiced initial fricatives is, however, obvious in words like FERN  Finally, to illustrate the difference in correlation with low central and low back vowels, we may cite the figures for FORTY. The voiceless:voiced ratio with a low central vowel \[a, a:, a, a:\] is 13:17~ while with a low back \[D, ~\] it is 27:18.</Paragraph>
    <Paragraph position="4"> It is clear that even more study, of individual words and individual localities, is needed before all the complications of this one dialect feature can be unraveled. We should, for example~ take into account the second and third responses for many of the words~ many of which were taken from incidental conversation and hence are inclined to be more natural. Even casual inspection of the data indicates that they show a much higher incidence of initial voicing than do the citation forms. But we hope that this paper has shown that, given adequate and convenient data, the computer can be of inestimable aid to the dialectologist.</Paragraph>
  </Section>
class="xml-element"></Paper>