Phonetic Change and Phonology Josef Fruehwald University of Pennsylvania May 21, 2011 MFM 19 in This Paper, I Will Attempt to Support the Following Two Argu- Ments

Home , Phonological change

Phonetic Change and Phonology Josef Fruehwald University of Pennsylvania May 21, 2011 MFM 19 In this paper, I will attempt to support the following two arguments

(1) Phonetic change is a distinct type of language change, display- ing qualitatively different dynamics from syntactic, morphological, and phonological change. (2) However, there is good evidence that categorical phonology plays a crucial role in the mediation of phonetic change.

I will propose that a model of phonetic change which adequately captures these effects, describes it as a change in language speciﬁc phonetic implementation of phonological objects. Phonetic change, then, can be localized to the implementation of particular phonological features, or bundle of features. I will conclude with some remarks about the consequences of this model for the nature of phonological representation.

Phonetic Change is Qualitatively Distinct from Other Language Change

When introducing the topic of language change, the fact that it typically follows an S-shaped curve is mentioned early on (Kroch, 1989; Labov, 1994, inter alia). However, there is a crucial difference between the S-shaped curve in most language changes, and the S-shaped curve in phonetic changes. In syntactic changes, the curve represents the frequency of use of the novel variant. The observations from syntactic change are categorically either the old, or innovative form, as 1.0 illustrated in Figure 1. 0.8 This categorical variation, combined with the logistic curve of change, led Kroch (1989, 1994) to treat syntactic change as competing 0.6 syntactic grammars. The canonical example is the rise of do–support, 0.4 Change which Kroch (1994) treats as competing variants of Tense. One vari- 0.2 + ant triggers V-to-T movement (T ) and the other does not (T−). By 0.0 hypothesis, speakers who acquired language at some point part way Time through the change would use both variants according to the frequency of use in the speech community at the time. An example Figure 1: Distribution of observations in syntactic, morphological, and phono- from Samuel Pepys is given in (3). logical change. phonetic change and phonology 2

(3) The Diary of Samuel Pepys, 1666–1667 (emphasis added) a. “He understands not the nature of the war.” b. “Fellows that did not understand them . . . ” This approach to variation shares much in common with variable rules (Cedergren and Sankoff, 1974), which incorporates the probabil- ity of application into the rule description itself. While the grammar contains probabilities in variable rule analyses, the outcomes of those rules are still categorical.

In phonetic changes, on the other hand, the S-shaped curve represents the quality of use of a linguistic form, as illustrated in Figure 0.8 2. The ﬁrst large scale acoustic study of phonetic change, Labov 0.6 et al. (1972), concluded that phonetic changes progress continuously through the phonetic space. That is, midway through a phonetic 0.4 Change change, speakers don’t produce a mixture of maximally innovative 0.2 forms and the old forms. Rather, they produce forms intermediate between the starting point and ending point of a change. Time Neither grammar competition nor variable rules are immediately Figure 2: Distribution of observations in extendable to describe continuous phonetic change. phonetic change.

Is phonetic change truly continuous?

Of course, it is necessary to demonstrate that phonetic change progresses continuously. If speakers were actually producing tokens drawn from highly overlapping bimodal distributions, and we were to simply calculate the mean of that mixture, it would produce the appearance of continuous change. I will take the raising of /ay/ before voiceless consonants (hence- forth (ay0)) in Philadelphia as a test case. This was originally iden- tiﬁed as a new and vigorous change led by men, (Labov, 1994), and has been linked to masculinity and toughness (Conn, 2005; Wagner, 2007). The raised variant has also started to appear in contexts where it was not previously conditioned (Fruehwald, 2007), but I will be fo- cusing exclusively on contexts which unambiguously trigger raising. See the section on the Philadelphia My data is drawn from the developing Philadelphia Neighborhood Neighborhood Corpus below for details about this data. Corpus (Labov and Rosenfelder, 2011). Figure 3 displays the mean F1 value (in normalized z-scores) for (ay0) for 47 Philadelphian speakers born between 1889 and 1991. We can observe a rapid and cohesive shift from a not-quite fully low position to a mid position in this data. However, these means don’t necessarily indicate that (ay0) is raising in a continuous change. In order to determine that, we have to decide whether for speakers midway through the change, their distribution of their tokens looks more like it has a single mode, or two. phonetic change and phonology 3

Figure 3: Change in (ay0) height in 0.0 Philadelphia by date of birth.

0.2

0.4

0.6 Sex f 0.8 m F1 z-score 1.0

1.2

1900 1920 1940 1960 1980 Date of Birth

That is, we have to determine if speakers have a single phonological object that is changing in its phonetic realization over time, or whether they have two phonological objects with ﬁxed phonetics (say, /5I/ and /@I/) which are shifting in frequency of use. If the situation is the latter, we can develop a hypothesis of what the the distribution of pure /5I/ and pure /@I/ should look like. Presumably, speakers with the lowest realizations of (ay0) should be closest to pure /5I/, and speakers with the highest realizations Mean F1 Sd should be closest to pure /@I/. So, I took the 4 speakers with the 5I 1.29 0.41 @I lowest (ay0), and calculated their average F1, and average standard 0.05 0.48 Table 1: Mean and standard deviations deviation for F1 to represent the distribution parameters of /5I/. I of z-scored F1 for the 4 most advanced, did likewise for the 4 speakers with the highest (ay0) to represent the and 4 most conservative speakers with distribution parameters of /@I/. These parameters are displayed in respect to (ay0) Table 1. Using these parameters, I simulated the change from /5I/ to /@I/ as changing mixture of the two (Figure 4). If the observed data from the change is sufﬁciently similar to the simulation, then a categorical -1 model of phonetic change could be reasonable. If it is not, then I would conclude that this change has progressed continuously. 0 F1 1 One clear consequence of the mixture model is that there should be a strict relationship between the estimated mean value of 2 a speaker, and the standard deviation of their distribution. The standard deviation of a mixture of two distributions will be larger than Time the the standard deviation of either one in isolation, and it should Figure 4: Simulated change in (ay0) be larger the more even the mix. Therefore, if the mixture model height based on mixing two distributions phonetic change and phonology 4

is correct, we should predict that the standard deviation of speakers’ distributions of (ay0) should reach a peak mid-way through the change. In fact, we can estimate what the s.d. of a speaker’s distribution ought to be based on the simulation. Figure 5 compares the relationship between the mean and s.d. from the simulation to the observed data.

Figure 5: Comparing the relationship

0.8 of the sample mean to sample standard deviation for observed speakers and simulations. Observed data in black. 0.7

0.6 sim obs

F1.sd 0.5 sim

0.4

0.3

0.0 0.2 0.4 0.6 0.8 1.0 1.2 F1

There may be a slight tendency for speakers midway through the (ay0) raising change to have larger standard deviations, but most speakers have the same standard deviation throughout all stages of the change. I conclude from this that phonetic change progresses as a 58 MCMILLAN, CORLEY, LICKLEY single target moving through the phonetic space, estimates of which For the first analysis we used a measure of tongue contact variability are represented in Figure 3 above. calculated relative to the reference articulation of the intended target onset. There was a marginal effectTP of Context; articulation tended to vary more from the intended target inq theqMM non-lexicalM context than in the mixed context: difference scores of 2.09qqq vs. 1.93, FM1M(1,M 6)4.81, p.071, MSE.004; F2(1, Is there any way that categorical variation could produce 47)3.44, p.070,qq MSE.201. NoM other effects were significant (all FsB1). + the appearance of continuous change? I will briefly consider two op- T ∼− VP The second analysis was based on theqMM tongue contact variability of each tions. First, perhaps it is possible that in cases of competition betweentarget calculated relative to theq referenceqq M articulationMM of the competitor onset. This analysis showed aq significantqq effect ofMM Competitor (marginal by- two categorical phonological objects can result in their blend in pro- items): Articulation was more similar to the competitor onsets when targets had a real word competitor. than . . when they did not: VP difference scores of 4.41 ×S duction. This blending would be possible for phonetic change, since vs. 4.55; F1(1, 6)15.71, p.007, MSE.0008; FM2(1, 47)3.62, p.063, qqq MMM MSE.226. No other effects were significantqq (all FsB1.07).MM something halfway between [5I] and [@I] is well defined, while for Figure 2 shows how tongue contactq variesq relative to theM intended and competitor reference articulations acrossV the experiment. Points... in the upper syntactic change, head movement half way to Tense is not. left corner would be likely to be perceptually transcribed as ‘correct’, and Some evidence from speech production research suggests that those in theFigure lower right6: cornerNot a as well ‘errors’. defined Importantly, syntactic there are also several points representingmovement articulations given competition which are intermediate on T. between the such blending may be possible. Utilizing a variant of the tongue- twister paradigm, McMillan et al. (2009) induced the probable trans- position of the onsets of nonsense words in their participants, leading to these onsets being in competition for production. Their main result, reproduced in Figure 7, was that while in most cases, speakers produced either fully canonical versions of the target segment (upper 2345

left quadrant) or fully canonical versions of the competitorDownloaded By: [University of Pennsylvania] At: 21:08 1 May 2009 (lower

right quadrant), they also produced quite a few intermediate forms. Distance from Competitor Reference 1

12345 Distance from Intended Reference Figure 2. ScatterplotFigure showing7: Gradient calculated tongue productions contact distance result- of each recorded onset articulation from the intended and competitor references. Points in the upper left corner represent articulationsing from similar categorical to the intended competition, reference but different from from the competitor reference; points in the lower right corner represent articulations similar to the competitor reference butMcMillan different from the et intended al. (2009 reference.) phonetic change and phonology 5

However, I don’t believe this result is extendable to explain con- either tinuous phonetic change. To begin with, it would reduce the ob- -2 served patterns in phonetic change to speech errors. Secondly, un- -1 blended categorical competition does take place. To choose a rel- 0

atively uninteresting example, Figure 8 plots all tokens of either F1 from the Philadelphia Neighborhood Corpus. Most speakers say 1 [i:DÄ], some say [aIDÄ], and [eIDÄ] is unattested. Finally, in the case of 2 (ay0), the allophone undergoing a phonetic change is the product of a 2 1 0 -1 phonological process. For this blending account to apply to (ay0), the F2 phonological process which produces (ay0) would have to produce Figure 8: All tokens if either from the two outputs simultaneously, which then blend in production. Philadelphia Neighborhood Corpus. No intermediate tokens are observed.

142 The Speech Community Another approach to treating phonetic change as categorical The Philadelphia Vowel System 143 tight cluster of mean values in that area, where (obr) is just barely /uwJl. The most remarkable advance is the raising and backing of F2 variation would be to either expand the feature set, or increase the 2600 2400 2200 2000 1800 1600 1400 1200 1000 800 For Rick Corcoran, this is a high back vowel. Since it is unrounded, it 400 not as back as (ohr) and /uwll, but it is not much lower. 111.1 has followeG the same trajectory. 450 uwe number of possible values a feature can take (Weinreich et al., 1968; oiyF ohr' 500 These individual vowels provide dramatic evidence of rapid sound i Oyo' across three generations, where vowel nuclei move from one extreme 550 oh.... Flemming, 2004). The major problem withthe vowel an space ton the-ary other. The approach chapters to follow will is assemble that the 600 values of 112 speakers in a normalized vowel space. Multiple regressiOd owF on those means will help to unravel the ways in which social class, gender,; 650 ...... in order to capture the necessary degrees of freedom involved in a °ahr age, and social network position are correlated with this advancement F1 700 change. Before proceeding, it may be helpful to add a dynamic phonetic change, the role features play in defining natural classes 750 ...j/. to this brief overview of the Philadelphia vowel system by simply cha1"tiM their mean values against age. 800 would be eroded. Taking Philadelphia, for example, given the num- 850 00\ ayV O 900 4.5 Development of sound changes in apparent time ------I ber of phonetic changes in progress in the vowel system, it would be 1P<.10 P<.05 P<.01 P<.001 The diagrams to follow trace the development in apparent time of ------Philadelphia sound changes as reflected in mean values of ten-year Figure 4.8 Mean values of all Philadelphia vowels with age coefficients for the Neighborhood Study. Circles: mean FI and F2 values. Heads of arrows: expected surprising if the natural class defined bygroups any in the givenPhiladelphia Neighborhood feature Study. and We cannot value of Figure 9: Active sound changes in values for speakers 25 years younger than the mean. Tails of arrows: expected average the unnormalized physical values of the combined sample of progressvalues for speakers in 25 Philadelphia,years older than the mean from Labov were larger than 1. women, and children, since differences in vocal tract length lead to differences in physical realization of the "same" sound. The mean (2001) to be shown below are normalized, using methods to be developed detail in the next chapter. In chapter 3 of volume 1, an overview of the Philadelphia sound cnanpl 2500 -r.------, in progress was provided by figure 3.6, reproduced here as figure 4.8. Conclusion mean values for 112 subjects of the Philadelphia Neighborhood 2400 are shown, and the age coefficients measuring movement in apparent are indicated by arrows whose length is proportional to the age coeffici_ 2300 The head of the arrow indicates the expected value of FI and F2 2200 -lc-oohN Phonetic change is qualitatively differentspeakers from 25 years younger other than kindsthe mean, and of the tail lan- indicates the -lc-oohS responding position for speakers 25 years older than the mean. lij 2100 -I:r- oohD Five of the vowel changes involve raising along the front diagonal of ""'-eyC guage change, and cannot be described in terms of competition, or 2000 vowel system. It will appear throughout that the physical index of ---- aw changes that is most sensitive to social variables is F2, rather than 1900 variation between categorical alternatives.Accordingly, Rather, figure 4.9 compares it is thebest mean values described of the second formanllj decade by decade, for the three allophones of lreh/: (rehN), (rehS), 1800 as the continuous movement of a phonetic(rehD), along target with the two through new and vigorous the changes, pho- (aw) and (eyC). Figure 4.9 shows a monotonic increase in values with decreasing Under 20 20-29 30-39 40-49 50+ with somewhat less regular progression for the variable with the least Age netic space. (rehD). As the individual vowel charts made clear, the most advanced I leave aside, here, the literature sug- the three /a:.h/ allophones is (rebN) in man, understand, etc. The other Mean second formants by lO-year age groups of five Philadelphia changes involving raising and frOnting Furthermore, speakers are clearly sensitive to fine grained pho- gesting that knowledge of the phonetic target in phonetic space affects not only netic differences, as they index important social and indexical infor- speaker’s production, but also phono- mation (Labov, 2001; Conn, 2005; Wagner, 2007; Eckert, 2010) without logical perception (Hay et al., 2006; De Decker, 2008; Kendall and Fridland, altering phonological identity. [f2It] and [f2fflIt] are not phonological 2010). minimal pairs, but you should be more cautious about someone who says [f2fflIt mi:].

Categorical phonology plays a role in phonetic change.

Despite the fact that phonetic change cannot be described as occur- ring at the level of categorical phonology, there is some evidence to suggest that it plays a crucial role as a mediator. Most importantly, given the regularity of sound change, the unit of phonetic change is typically deﬁned as the phonological segment. phonetic change and phonology 6

Labov (2010, Ch 13) walks through a number of case studies of phonetic change in North America, and concludes that most apparent lexical effects are explicable by phonetic context. Phonetic change affects all members of a higher organizational unit. True cases of lexical diffusion seem to occur not at the lower levels of phonetic implementation of these units, but rather, at the higher organizational level, thus are outside the purview of discussion here. Additionally, it is the case that at all stages of a phonetic change, the sound system as a whole remains analyzable as a system of equivalencies and contrasts. That is, if a language undergoes a change from stage A to stage B, all intermediate stages are also a language. This is a non-trivial fact, and can be contrasted with the exemplar based modeling of de Boer (2001), where languages pass through “brief period[s] of chaos” from one stage to another.

Parallel chain shifts are also important evidence that the unit of phonetic change may be even smaller than a segment, encompassing a phonological natural class. For example, there is the trend across uw ow North America to front the long, back upgliding vowels, /uw/, ow 0.73 /ow/ and /aw/. Looking at the mean values from the Atlas of North aw 0.48 0.57 Table 2: Pearson correlation coefﬁcients American English (Labov et al., 2006), the backness of these three for /uw/, /ow/ and /aw/ backness vowels are more highly correlated with each other than with any across North America (Labov et al., other vowels. Table 2 displays a correlation matrix containing Pear- 2006) son correlation coefﬁcients, and Figure 10 plots these relationships.

/aw/ vs /ow/ /aw/ vs /uw/ /ow/ vs /uw/

1000 1000 1000

1200 1200 1200

1400 1400 1400

ow 1600 uw 1600 uw 1600

1800 1800 1800

2000 2000 2000

2200 2200 2200

2200 2000 1800 1600 1400 1200 1000 2200 2000 1800 1600 1400 1200 1000 2200 2000 1800 1600 1400 1200 1000 aw aw ow

Figure 10: The correlation of /uw/, Given the high correlation of phonetic backness between these /ow/ and /aw backness across North America (Labov et al., 2006) vowels, it is tempting to say these are all realizations of a single phonetic change which has generalized to a phonological natural class. An even more important case is the Canadian Shift, where the precipitating cause affects only one element of the natural class undergoing the shift. First described by Clarke et al. (1995), the Cana- dian shift is presumed to be triggered by the merger of /a/ to /O:/. on this shift, said that /E/ and /I/ lowered as a result of the merger (see Figure 14), and this was also the description of the shift in the ANAE.

E O

æa

The Canadian Shift was firstFigure described 14: as Canadian being more Chain accurate Shift ly called a parallel shift by Boberg (2005), and nearly every subsequent study in various dialects has followed suit. The gapThis filling kind behavior of chain of shift/æ/ is is a amenable classic phonetic to a gap reaction creation,andeasilyexplainedwithin and filling usage based model. an exemplarHowever, theoretic the Canadianphonetic framework. Shift change does not and just take phonology the form of a 7 chain shift. Boberg (2005) However,found that the in subsequent Montreal, the retraction effect of of /æ/ /e/ retraction and /i/ cannot on thebe other explained short vowels as being is parallel a similarretraction gap filling of / reaction.E/ and /I Instead,/. Durian it seems (2009) as reports if what the sta samerted o kindff as aof phonetic parallel retractionpressure in on /æ/Columbus has generalized Ohio. to the phonetic implementation of [ back]. This isn’t a change Clarke et al. (1995) described the following steps of the shift as the − in the phonological status of [ back], but rather in the phonetic implementation function − I retraction of /æ/, followed by the lowering of /Emediated/ and /I by/. [ How-back]. − ever, recent investigations in Montreal (Boberg, 2005Ithinkitisnecessary,giventherelationshipofelementsi) and Columbus, ntheCanadianShift,to OH (Durian, 2009) have described the stages followingsay that the phonemes low-back as atoms can have some particularized phonetic realization. That is, the gap in the low-back space triggered /æ/E as a whole to becomeretracted.Then,this merger instead as parallel retraction of /æ/, /E/ and /I/. retraction generalized out to the other members of the naturO al class /æ/ belongs to, which Regardless of the particular ordering, or the definitionis defined along of neces- the dimension along which /æ/ is moving. That is, the retraction of /æ/ sary or sufficient events (Bigham, 2009), clearly thewas only reanalyzed vowel as on being partially due to theæa implementation function for [ back]. − which there is a relevant phonetic pressure for retraction is /æ/. Figure 11: Canadian Parallel Shift. The generalization of retraction to /E/ and /I/ occurs4.3The Spirantization exactly Canadian along Shift was firstFigure described 15: Canadianas being more Parallel accurate Shiftly called a parallel shift lines we would expect to be defined by a phonologicalForby Boberg some natural hypothetical (2005), class and language:nearly every subsequent study in various dialects has followed suit. The gapWhat filling is behavior clear is that of /æ/ the is phonetic a classic motivation phonetic reaction for either,andeasilyexplainedwithin form of the Canadian Shift acts (specifically [-tense, -back]). Describing this generalization merely an exemplaruponk /æ/.x theoretic Whether framework. it is due to the phonetic gap created by merger, or a consequence as “analogy” calls for a principled way to delimit the•However,of elements listening→ the and for subsequent speaking retractionto merged ofspeakers /e/ and in one’s/i/ cannot speech be community, explained as as suggested being a by Bighamt T (2009), only /æ/ is directly affected. In the parallel forms of the Canadian Shift, which the analogy extends to ({æ, E, I}) from thosesimilar which• → gap it filling does reaction. not Instead, it seems as if what started off as a phonetic pressure on /æ/however, has generalized the samephonetic to the phonetic change implementation of backing happens of [ toback]. /E/ and This / isn’tI/. Durian a change (2009) ({2, U}). My suspicion is that any such principled attemptp wouldf look − in the•called phonological→ this generalization status of [ aback], classic but four-part rather analogy: in the phonetic implementation function very similar to [-tense, -back]. − mediated(19) by(a) [ æ:æback].:: :X(= ) The change of− [k] ffl[x] andE [t] Effl[T]bothinvolveundershootofthetonguetoreachastop Ithinkitisnecessary,giventherelationshipofelementsi→ → ntheCanadianShift,to target. The(b) spreadE : Effl of:: thisI :X(= spirantizationfflI) to [p], a bilabial stop, involves spreading along a Phonetic change in language specific phoneticssaymore that abstract phonemes dimension, as atoms of something can have somelike [ particularizedcontinuant] phonetic realization. That is, the gapWhile in it the is clear low-back that spacethe retraction triggered of /æ/ /E/− and as a /I whole/ is a generalization to becomeretracted.Then,this of the phonetic pressure retractionon /æ/ generalized of some sort, out towithout the other accounting members for of the the phonological natural class relationship /æ/ belongs between to, which these Based on the arguments in the previous two sections,is defined I propose along the that dimension along which /æ/ is moving. That is, the retraction of /æ/ the best way to account for both the continuousnesswas reanalyzed of phonetic as being partially due to the implementation25 function for [ back]. − change, and the mediating role of phonology, is to treat phonetic change as changing phonetic implementation of4.3 (relatively Spirantization stable) 4 phonological representations. This approach capturesFor some both hypothetical the con- language: tinuous and categorical properties of phonetic change. First, phonetic k x implementation involves the computation of targets• in→ continuous Phonology t T phonetic space. Changes in this computation will produce• → different targets in this phonetic space. Second, the output of thep computationf is strictly dependent on its input, which is (by hypothesis)• → categorical Surface Phonological Representation The change of [k] [x] and [t] [T]bothinvolveundershootofthetonguetoreachastop phonological objects. → → target. The spread of this spirantization to [p], a bilabial stop, involves spreading along a The model of the phonology-phonetics interface I’m adopting here more abstract dimension, of somethingPhonetic like Implementation [ continuant] Rules is similar to “generative phonetics.” Figure 12 is a slightly modified − version of system proposed in Keating (1990). Phonetic Target Representation (4) Phonology categorical categorical → Phonetic Alignment4 Constraints (5) Phonetic Implementation Rules categorical continuous → Output for Articulation (6) Phonetic Alignment Constraints (Cohn, 1993; Zsiga, 2000) Figure 12: The Phonology-Phonetics continuous continuous Interface → The substantive content of my proposal is that Phonetic Implemen- tation Rules are part of a speaker’s non-arbitrary, learned knowledge of their language (Pierrehumbert, 1990; Kinston and Diehl, 1994). I believe this is a necessary assumption, given the available data on phonetic change. phonetic change and phonology 8

What are the units that the phonetic implementation rules operate over? Arguing strictly from data on phonetic change, I believe there are two answers.

(7) Phonological Features (8) Surface Segments

Features

Clearly, there are many other reasons to presume that phonetic implementation rules operate over phonological features. For instance, if segments didn’t share phonetic properties with other segments with which they also share phonological features, how could their featural identities be learned? On the basis of data from phonetic change, parallel shifts are also evidence that phonetic implementation rules operate over features. If phonetic change takes place in phonetic implementation, as I am arguing, and if a phonetic change is affecting an entire natural class, as is the case in the Canadian Shift, then the process of phonetic implementation undergoing the change must be one that makes reference to the natural class.

Surface Segments

When looking at phonetic change, there are some cases where it appears that one segment is changing, but not any of the other elements in its natural class. For example, (ay0) is raising and backing in Philadelphia, but no other segment it might share a natural class with is. Perhaps this change could be connected to the fronting and raising of /aw/, or /ey/ in non-ﬁnal position, but that may be a stretch.

Words?

As I discussed above, Labov (2010) found that for the most part, apparently lexical effects on phonetic change are the product of contextual effects. I would conclude, then, that phonetic change does not provide strong evidence for word speciﬁc phonetics (Pierrehum- bert, 2002). Language changes where clear lexical effects have been identiﬁed can be described at a higher level of abstraction than the phonetics.

Phonological Reasoning about Phonetic Change

This approach to reasoning about phonetic changes can also enrich phonological investigation. Take, for example, the phonetic change in /ow/ in the American South (Labov et al., 2006). Figure 13 plots the phonetic change and phonology 9

fronting of /ow/ before stops, and the retraction of /ow/ before /l/.

Figure 13: Phonetic change of /ow/ before stops and /l/ in the American 80 South 70

60 Manner 50 l Age 40 stop 30

0.0 -0.5 -1.0 -1.5 F2

Under my account of phonetic change, there are two possible reasons for why /ow/ is shifting in two different directions. (9) a. There has been a phonetic change in /l/, causing it to have a different contextual effect on /ow/. b. /owT/ and /owL/ have different surface phonological representations, and the phonetic implementations of these representations are changing independantly. If (9b) is the cause, then there are two possible ways for /owT/ and /owL/ to appear different in their surface phonology. (10) a. There has been a phonemic split between /owT/ and /owL/, so that they now have different underlying forms. b. There is an active synchronic phonological process which changes /ow/ in the context of a following /l/. Further empirical investigation would be necessary to determine whether or not there has been a relevant phonetic change in /l/. Alternatively, if there is a phonological difference between /owT/ and /owL/, further investigation would be necessary to determine what the difference is, and whether or not it’s the product of an active phonological process. To be sure, my proposal for phonetic change does not answer these questions, but it does a lot to deﬁne the relevant questions for future research.

Another similar case is the split between /ey/ in word ﬁnal position and /ey/ in all other positions in Philadelphia. Figure 14 plots the data from the Philadelphia Neighborhood Corpus. phonetic change and phonology 10

Figure 14: Phonetic change in Philadel- phia affecting /ey/ in non-ﬁnal position. -0.6

-0.4 VClass -0.2 ey F1 eyF 0.0

0.2

1900 1920 1940 1960 1980 Date of Birth

In this case, I believe the only possible explanation for this split is different surface phonological representations between /ey/ and /ey#/, allowing /ey/ to undergo a phonetic change independently from /ey#/. Again, there are two possibilities for what may cause this difference.

(11) a. There has been a phonemic split between /ey/ and /ey#/, so that they now have different underlying forms. b. There is an active synchronic phonological process which changes /ey/ in word ﬁnal contexts. -1.0

In this case, I believe there is sufﬁcient data to suggest that (11b) is -0.5

the most probable explanation. Figure 15 illustrates that for most 0.0

speakers born after 1945, there is a consistent difference between F1 0.5 day and days in the expected direction. Given the direction of the phonetic difference between /ey/ and /ey#/, an initial formulation 1.0 of the phonological process which differentiates them is given in 12. 2.0 1.5 1.0 0.5 0.0 (12) ey [ peripheral] / # F2 → − Of course, this description of the situation implies that there were Figure 15: Comparing days (points) to day actually two changes which occurred, necessarily in the following (arrowheads) in Philadelphians born after 1945. order.

(13) a. A phonological change, introducing the process of /ey/ laxing in word ﬁnal position. b. A phonetic change raising tense /ey/. phonetic change and phonology 11

Conclusion

The proposal I’ve put forth here is intended to be an initial sketch of a framework within which to reason about sound change. Phonetic change has unique dynamics from language change in other do- mains, thus requires a specialized change model. Further work in this vein should attempt to provide descriptions of the phonetic implementation rules, descriptions of the phonetic target representations, and an account of any feedback relationship between the phonetic implementation of features, and those features’ contrastive roles (i.e. dispersion (Liljencrants and Lindblom, 1972) or margins of security (Martinet, 1952)). My proposal also implies that phonological representations are “substance free” (Odden, 2006; Blaho, 2008), since the phonetic implementation of particular phonological features are language spe- ciﬁc, and learned. If the role of phonological features are not to de- ﬁne the phonetics of a segment (it’s the implementation rules which do that), then their purpose is strictly formal.

Philadelphia Neighborhood Corpus Data

This data is drawn from transcribed interviews with native Philadelphians, which were conducted between 1973 and 2010. The transcriptions were time aligned to the audio at the word and phone level using a mod- iﬁed version of the Penn Phonetics Lab Forced Aligner (Yuan and Liberman, 2008). Formant measurements were then automatically measured (Evanini, 2009; Labov and Rosenfelder, 2010).

References

Bigham, Douglas S. 2009. Correlation of the Low-Back Vowel Merger and Trap-Retraction. In Penn Working Papers in Linguistics, ed. Kyle Gorman and Laurel MacKenzie, volume 15.2. Blaho, Sylvia. 2008. The syntax of phonology: A radically substance free approach. Doctoral Dissertation, University of Tromsø. Boberg, Charles. 2005. The Canadian shift in Montreal. Language Variation and Change 17:133–154. Cedergren, Henrietta J., and David Sankoff. 1974. Variable rules: Performance as a statistical reﬂection of competence. Language 50:333–355. Clarke, Sandra, Ford Elms, and Amani Youssef. 1995. The Third Dialect of English: Some Canadian evidence. Language Variation and Change 7:209–228. Cohn, Abigail. 1993. Nasalisation in English: Phonology or Phonetics. Phonology 10:43–81. Conn, Jeffery. 2005. Of ‘moice’ and men: The evolution of a male-led sound change. Doctoral Dissertation, University of Pennsylvania. de Boer, Bart. 2001. The Origins of Vowel Systems. New York: Oxford University Press. phonetic change and phonology 12

De Decker, Paul. 2008. Gender and age differences in phonetic categorization and evaluation of vowel stimuli during a change in progress. Poster presented at NWAV37, October 2008. Durian, David. 2009. Purely a chain shift?: An exploration of “Canadian Shift” in the US Midland. Paper presented at NWAV 38. Eckert, Penelope. 2010. Affect, sound symbolism, and variation. In Penn Working Papers in Linguistics, ed. Kyle Gorman and Laurel MacKenzie, volume 15.2, 70–80. Evanini, Keelan. 2009. The permeability of dialect boundaries: A case study of the region surrounding Erie, Pennsylvania. Doctoral Dissertation, University of Pennsylvania. Flemming, Edward. 2004. Contrast and preceptual distinctiveness. In Phonetically Based Phonology, ed. Bruce Hayes, Robert Kirchner, and Donca Steriade, 232–276. Cambridge University Press. Fruehwald, Josef. 2007. The spread of raising: Opacity, lexicalization, and diffusion. CUREJ: College Undergraduate Research Electronic Journal URL http://repository.upenn.edu/curej/73. Hay, Jen, Paul Warren, and Katie Drager. 2006. Factors inﬂuencing speech perception in the context of a merger-in-progress. Journal of Phonetics 34:458–484. Keating, Patricia. 1990. Phonetic representations in a generative grammar. Journal of Phonetics 18:321–334. Kendall, Tyler, and Valerie Fridland. 2010. Mapping production and perception in regional vowel shifts. Penn Working Papers in Linguistics 16. Kinston, John, and Randy L. Diehl. 1994. Phonetic knowledge. Language 70:419–454. Kroch, Anthony. 1989. Reﬂexes of grammar in patterns of language change. Language Variation and Change 1:199–244. Kroch, Anthony. 1994. Morphosyntactic variation. In Papers from the 30th Regional Meeting of the Chicago Linguistics Society: Parasession on Variation and Linguistic Theory, ed. K. Beals. Labov, William. 1994. Principles of linguistic change. Volume 1: Internal Factors. Blackwell. Labov, William. 2001. Principles of linguistic change. Volume 2: Social Factors. Language in Society. Oxford: Blackwell. Labov, William. 2010. Principles of Linguistic Change. Volume 3: Cognitive and Cultural Factors. Blackwell. Labov, William, Sherry Ash, and Charles Boberg. 2006. The Atlas of North American English. Phonetics, Phonology and Sound Change.. Berlin: Mouton de Gruyter. Labov, William, and Ingrid Rosenfelder. 2010. PLOTNIK/PRAAT Interaction and Automatic Vowel Anal- ysis. Workshop at NWAV 39. Labov, William, and Ingrid Rosenfelder. 2011. The Philadelphia Neighborhood Corpus. Labov, William, Malcah Yaeger, and Richard Steiner. 1972. A Quantitative Study of Sound Change in Progress. The U.S. National Survey. Liljencrants, Johan, and Björn Lindblom. 1972. Numerical Simulation of Vowel Quality Systems: The Role of Perceptual Contrast. Language 48:839–862. Martinet, André. 1952. Function, structure and sound change. Word 8:1–32. McMillan, Corey, Martin Corley, and Robin J. Lickley. 2009. Articulatory evidence for feedback and competition in speech production. Language and Cognitive Processes 24:44–66. phonetic change and phonology 13

Odden, David. 2006. Phonology ex nihilo. Paper presented at the the Tromsø Phonology Project Group Meeting. Pierrehumbert, Janet B. 1990. Phonological and phonetic representation. Journal of Phoneticsn 18:375–394. Pierrehumbert, Janet B. 2002. Word-specific phonetics. In Laboratory Phonology VII, 101–139. Mouton de Gruyter. Wagner, Suzanne E. 2007. We act like girls and we don’t act like men: The use of the male-associated variable (ay0) in South Philadelphia. In Penn Working Papers in Linguistics, ed. Tatjana Scheffler, Joshua Tauberer, Aviad Eilam, and Laia Mayol, volume 13.1, 393–406. Philadelphia: University of Pennsylvania Press. Weinreich, Uriel, William Labov, and Marvin Herzog. 1968. Empirical foundations for a theory of language change. In Directions for Historical Linguistics, ed. W. Lehmann and Y. Malkiel. U. of Texas Press. Yuan, Jiahong, and Mark Liberman. 2008. Speaker identification on the SCOTUS corpus. In Proceedings of Acoustics. Zsiga, Elizabeth C. 2000. Phonetic alignment constraints: consonant overlap and palatalization in english and russian. Journal of Phonetics 28:69–102.