Against Gradual Phonologization

Josef Fruehwald, University of Edinburgh
November 16, 2013

Introduction

Welcome to my Dissertation Defense 2.0 1.1

I'll be making the case that the conventional wisdom surrounding the transition of phenomena from phonetic to phonological is not that well supported by studying language change in progress.

I'll argue that phonological reanalysis occurs at the onset of sound changes, not as late stage reanlyses.

In the process I'll also be trying to make the case that the difference between "phonetic" and "phonological" is not just a matter of degree, but of quality.

Terminological Issues

Terminological Issues

Hyman (1976)

Phonologization is the reanalysis of an "intrinsic" phonetic property (universal, physiological, etc.) as an "extrinsic" phonetic property (language specific, under speaker control).

After phonologization comes phonemicization.

Bermúdez-Otero (2007)

Phonologization is the reanalyzis of an "intrinsic" phonetic property as an "extrinsic" phonetic property.

After phonologization comes stablization, where the extrinsic phonetic property becomes categorical.

Terminological Issues

Kiparsky (2013)

Phonologization is the reanalysis of phonological allophones (categorically conditioned) as phonemes or quasiphonemes.

Fruehwald (2013)

Phonologization is the introduction of a new phonological process into the grammar.

Terminological issues

Some work:

What I mean:

The PNC

Philadelphia Row Homes, 1973  
Philadelphia Row Homes, 1973
North Philadelphia Row Homes, 1973
North Philadelphia Row Homes, 1973
North Philadelphia Row Homes, 1973
North Philadelphia Row Homes, 1973
South Philadelphia Row Homes, 2012
South Philadelphia Row Homes, 2012
South Philadelphia Row Homes, 2012  
South Philadelphia Row Homes, 2012

The PNC

Automating Acoustic Analysis

Much of the time spent on measurement consists of locating the words of interest and storing these segments. More than one member of our research staff has projected a program for automatic location, segmentation, and measurement of vowel nuclei, but so far, all such attempts have led to an increase in gross error rates of several orders of magnitude. At present, we find there is no effective substitute for the careful examination and measurement of the formant trajectories of each individual vowel token by an analyst relying on both auditory and visual information, double-checking the computer’s analysis against auditory impressions.

Labov, Ash & Boberg (2006)

The PNC

Step 1: Transcription

The PNC

Step 2: Forced Alignment

The PNC

Step 3: Formant Analysis

plot of chunk means_plot 

The PNC

FAVE online suite

Large Volumes of Data

The Use of Change in Progress

When we observe a language change in progress, the change is taking place in speakers' knowledge of their language from generation to generation.

To know which aspect of their knowledge is changing, we really need to observe the sound change in progress.

The Use of Change in Progress

Metathesis

  1. It could be lexically gradual, moving one word at a time, possibly due to perceptual confusion (Blevins & Garret 2004)
    • e.g. chipotle -> chipolte
  2. It could be due to a new phonological process being added to the grammar, which produces phonetically abrupt variation.
  3. It could even be phonetically gradual
    • Andalusian pre-aspiration becoming post aspiration (Parrel, 2012; Ruch 2012)
      • ht > hth > th

In all three cases, the outcome is the same, but result from changes to three very different aspects of linguistic knowledge.

Phonological innovation

Tucker (1944) on the Philadelphia dialect (emphasis mine).

Both the [aɪ]-type diphthong and the [aʊ]-type diphthong exist in only one quality, whereas in most American dialects the first element is shortened and modified in quality before a voiceless consonant [...] No such distinction is made in the Philadelphia dialect.

Phonological Innovation

Today:

'Raymond', who was interviewed in 2010, was 19 (DOB 1981).

"I grew up here my whole life."
"Um, right now I'm undecided."
"life"
"undecided."

Phonological Innovation

It looks like between Tucker's description, and Raymond's birth, there was a phonological innovation in Philadelphia.

ay → -low/ __[-voice]
/ay/ - high
+ low
+ back
- round

/ay/ - high
- low
+ back
- round

Phonological Innovation?

Phonological Innovation?

plot of chunk ay_plot

Phonological Innovation?

plot of chunk ay_plot_2

Phonological Innovation?

plot of chunk plot_sim_sd

Phonological Innovation?

Continuous Change

It looks like /ay/ rose continuously across the 20th century.

There isn't any point in time where there is an obvious discontinuity, either across the community or within speakers, where there was a transition from a phonetic change to a phonological one.

Conclusion?

Is there any meaningful distinction to be made between a phonetic change and a phonological one?

Error Accumulation

Ohala (1981)

Error Accumulation

Prediction

Regular articulatory and perceptual biases will lead to gradual phonetic shift, and perhaps later stage phonological or phonemic reanalysis.

Non-phonologization

plot of chunk aw_basic_plot

Non-phonologization

plot of chunk aw_cohen_plot

Phonologization and Non-Phonologization

plot of chunk ay_aw_comp_plot

Phonologization and Non-Phonologization

In comparing these two phonetic changes, we can see that phonologization is not the deterministic outcome of a reliable and large phonetic phenomenon.

This result is even more striking when looking at the phonetic precursors of /ay/ alone

Mismatches

The model of gradual reanalysis would predict that phonetic changes should progress from the most favoring environment to the least.

In some cases, there is a mismatch between contexts which phonetically favor a change, and the eventual phonological conditioning in its outcome.

Mismatches

/ay/ raising

(Joos, 1942; Chambers, 1973; Moreton & Thomas, 2007)

Mismatches

plot of chunk early_ay

Faithful /d/ (235 ms) > Flapped /d/ (156 ms); Faithful /t/ (143 ms); Flapped /t/ (113 ms)

Mismatches

plot of chunk ingrid_plot

Data from Rosenfelder (2005)

Mismatches

Depending on what the phonetic precursor is, we'd expect to see one of these two patterns of sound change initiation and reanalysis.

plot of chunk plot_dur_rean

plot of chunk plot_glide_rean

Mismatches

It looks like /ay/ raising differentiates along phonological, not phonetic dimensions.

plot of chunk plot_flaps

Mismatches

I constructed a statistical model defining these relationships between /ay/ in these different contexts, utilizing b-splines (a non-linear curve fitting method) to model the effects over date of birth. Also included were random intercepts for speaker and for word.

The model was implemented in Stan, which used Hamiltonion Monte Carlo to estimate the effects. All priors were non-informative

Question: When, across the time course of the change, is /ay/ raising conditioned by phonological voicing?

Mismatches

Answer: At the very onset of the change.

/ay/ differentiates between pre-/t/ and pre-/d/ contexts in the exact same way whether the /t/ and /d/ are faithful, or flapped.

plot of chunk plot_diffs

Mismatches

Counter-proposal: Analogy

Most of the /ay/ followed by flapped /t/ were from morphologically complex words (e.g writer)

Maybe the immediate participation of these contexts in /ay/ raising was analogy with the morphologically simplex forms (write).

Response, Part 1

Well, the analogy appears to have immediately spread to all of the relevant cases all at once.

Mismatches

Response, Part 2

plot of chunk day_plot

Mismatches

/ey/ raising

plot of chunk plot_ey_raising

Mismatches

lmer(Diag ~ Decade*FolSeg + (FolSeg | Speaker) + (1|Word))

Decade is date of birth minus 1900 divided by 10.

C Estimates Interactions
Intercept 0.67 t = 17.02 -0.59 t = -4.83 FolSeg = V
0.28 t = 2.53 FolSeg = /l/
Decade 0.12 t = 15.56 -0.12 t = -6.35 FolSeg = V × Decade
-0.08 t = -4.12 FolSeg = /l/ × Decade

Mismatches

Wrapup

Phonetic favoribility is a weak predictor of which variants will undergo a change.

First is not fastest.

Phonological conditioning is present at the onset of these changes.

Conclusions and Analysis

The conventional wisdom regarding the gradual phonological reanalysis of phonetic effects:

  1. Either doesn't happen, or happens too rarely to be observed in the sound changes I've analyzed.
  2. Can't account for a number of the patterns in sound changes that I've observed.

Conclusions and Analysis

How do we resolve the the apparently early phonological differentiation with the gradual phonetic change?

Conclusions and Analysis

How do we resolve the the apparently early phonological differentiation with the gradual phonetic change?

Conclusions and Analysis

How do we resolve the the apparently early phonological differentiation with the gradual phonetic change?

Conclusions and Analysis

I have observed that when there is clear phonological conditioning of a sound change towards its end, that phonological conditioning is also in place at the sound change's onset.

Proposal

New phonological processes enter the grammar as strictly phonological innovations, not reanalyses of phonetic processes, and have relatively small phonetic correlates at first. This is similar to the "Big Bang" theory of Joseph & Janda.

Conclusions and Analysis

Phonological Conditioning

These phonological processes must have been present in the grammar either before or at the onset of the phonetic changes.

  1. ay → -low/ __ [-voice]
  2. ey → +peripheral/ __ C[-lat]

Is this Plausible?

There are plenty of examples children briefly entertaining phonological hypotheses which are markedly absent from the adult grammar, like Consonant Harmony.

Word Ahmal
duck g̊ʌk
kiss g̊ik
stop d̥ɔp ~b̥ɔp

(Smith (1973), quoted in Goad (1997))

Is this Plausible?

Why aren't we in phonological chaos?

Learning, both linguistic and social, could iron out a lot of idiosyncratic variation.

Is this Plausible?

Why aren't we in phonological chaos?

How do we know we aren't?

Mielke, Baker, Archangeli (forthcoming)

Conclusions

Future questions

Conclusions

Thanks

My committee

Bill Labov, Gene Buckley, Charles Yang, Ricardo Bermúz-Otero.

Many many helpful, and argumentative people

Audiences at NAPhC, the mfm, NELS, NWAV

The Penn Linguistics Deparment writ large

Especially denizens of the Linguistics lab.

References

References