Transcript "Frequency Effects in the Lexical Diffusion of Phonological Change"

"Frequency Effects in the Lexical
Diffusion of Phonological
Change"
Betty S. Phillips, Indiana State U.
[email protected]
LSA Summer Institute Workshop on
Variation, Gradience, and Frequency in
Phonology, 6-8 July 2007, Stanford U.
Lexical Diffusion
Definition:
Chen and Wang (1975): “a
phonological rule gradually extends
its scope of operation to a larger and
larger portion of the lexicon, until all
relevant items have been transformed
by the process.”
That lexical diffusion is sporadic
That it always affects the most frequent
words first.
That lexically diffused changes show no
clear phonetic conditioning
That phonetically regular changes are
productive, whereas lexically diffused
changes are not.
That lexical diffusion affects only
phonetically abrupt changes associated
with lexical rather than postlexical rules
That word frequency is independent of
word class.
That analogy and borrowing suffice to
account for lexical diffusion
That age of acquisition or discourse factors
account for patterns of lexical diffusion.
That frequency effects are found over a
population of speakers but not within
individual speakers.
That either lexical diffusion is the diffusion
of a completed sound change OR "all
phonological change starts with lexical
diffusion and most ends up
Neogrammarian, given enough time"
Misunderstanding #1
That lexical diffusion is "sporadic"
E.g., Hinskens (1998: 169): "Although variable by
nature, Neogrammarian sound change is preeminently 'systematic and recurrent' i.e. regular,
as there are no lexical exceptions. Lexically
diffuse and lexicalized sound change bring about
'non-systematic but recurrent' facts (Lloret 1997),
which are only partially regular, hence 'sporadic'."
One clear way in which lexical diffusion
is not at all random:
Word Frequency Effects
"Rarely-used words drag behind; very
frequently used ones hurry ahead.
Exceptions to the sound laws are formed
in both groups." -- Schuchardt (1885: 58)
“A cursory glance at the newspapers
suggests that adultery is on the increase in
this century. If you think slavery has been
abolished, go and look at the factory at the
end of the road. Every mother will tell you
that nursery schools are a mixed
blessing.” (Aitchison 2003)
Which retain the medial schwa, and which
lose it?
My ISU class project results:
Word
every
Celex - raw Schwa-less Schwa-ful
frequency
9788
22
4
factory
1064
18
8
nursery
580
19
7
slavery
134
10
16
adultery
107
15
11
cursory
32
9
17
Modern English final /t,d/ deletion
(Bybee 2002)
Env.
-ld
More likely to delete Less likely to delete
(Celex frequency)
(Celex frequency)
told (1763)
held (765)
-lt
felt (1449)
built (456)
-nt
sent (551)
meant (515), lent (25)
-pt
kept (750)
slept (120)
-ft/-st left (1503)
lost (759)
Stress shift in -ate verbs
Stress patterns C$C(C)-ate disyllables in Gimson's
1988 [1991] English Pronouncing Dictionary
Frequency Conservative Variable Innovative
⁄ —
— ⁄
(Celex)
0-7
X
filtrate...
9-36
X
mandate...
39-666
X
dictate...
Misunderstanding #2
That lexically diffused changes always
affect the most frequent words first.
E.g., Krug (2003): “It has become a
linguistic commonplace that highfrequency words and constructions tend to
lead phonological change. . . .”
Middle English Ormulum of /ø:/ > /e:/
(Phillips 1984: 328)--Verbs
Freq.
2-10
Example
Ave. % «e»
69
69
beodeÞÞ, nedenn,
chesenn...
cneow, fleon,
heold, ...
seon 'see'
355
beon 'be'
41
11-47
68
52
Middle English Ormulum of /ø:/ > /e:/
(Phillips 1984: 328)--Nouns
Freq.
1-8
Example
Ave. % «e»
49
21-27
breostlin, fend,
wheol . . .
deor, leom, treo, ...
68-82
preost, Þeod
4
deofell
1
158
6
Middle English Ormulum of short /ø/ > /e/
(Phillips 1984: 328)--Nouns
Freq.
Example
Ave. % «e»
1-2
berrhless, dorrf,
sede . . .
75
26-90
weorrc, weorelld,
heorrte
46
heofenn
6
154
1st example: Stress Shift x / > / x
UNCHANGED NOUNS
exchange
excuse
exempt
exhaust
express
CHANGED NOUNS
excise
exploit
export
extract
Ave. Freq. = 22.0
Ave. Freq. = 2.5
Summary: Development of Noun-Verb
Homographs (Phillips 2006)
40
35
30
25
20
15
10
5
0
Unshifted
co
r
n- e /c
om
-
a-
pr
ede
-
es
-
ex
-
di
s
-
Diatonic
Misunderstanding #3
That lexically diffused changes show no
clear phonetic conditioning (Kiparsky
1995).
Blevins (2004: 260): "lexical diffusion
(without clear phonetic conditioning) is the
diffusion of a completed sound change.”
Example: Southern Am. Eng.
Glide Deletion (Phillips 1994)
Freq.
Words
% Glideless
age 66-88
% Glideless
age 13-19
0-1
nude, Tudor, tuber, tunic,
dues, neutron, duly, tuba,
dude
59.4
69.8
1-10
nutrient, tutor, duel, duke,
durable, tulip, dune,
nuisance, neutral, nucleus
39.3
73.3
11100
Tuesday, numerous, tune,
duty, numeral, due, tube
19.0
54.1
1011000
knew, during, new
11.3
53.0
Misunderstanding #4
That phonetically regular changes are
productive, whereas lexically diffused
changes are not
Bakken 2001: " A phonetically regular
process is characterized by general
productivity, and such productivity seems
to be incompatible with lexical exceptions,
i.e. lexical diffusion. Lexical diffusion and
phonetic regularity are therefore truly
mutually exclusive.”
Labov, Yaeger, and Steiner (1972)
“We find no strong evidence for lexical diffusion in
the (æh) patterns of Detroit and Buffalo and
Chicago.”. . . .
Despite some initial oscillations the (æh) word
class seems to move upward as a whole, with
fine phonetic conditioning in the process. There
is some indication that the word mad is lower
than its phonetic class would justify for
several speakers. . . . the low position of mad
as compared to bad, ads, etc. , seems to be
lexically determined. " (93 – my bold).
Misunderstanding #5
That lexical diffusion affects only
phonetically abrupt changes
associated with lexical rather than
postlexical rules (Labov 1994;
Kiparsky 1995).
Reduction across word boundaries
(By definition "post-lexical")
Alba (2003): la 'the' or una 'a' + vowel-initial nouns
(e.g., iglesia 'church', hija 'daughter') in 20 hours
of tape-recorded interviews with 20 native
speakers of Spanish.
Strings with high ratio frequency underwent hiatus
resolution 87 per cent of the time, compared to
48 per cent for strings with low ratio frequency.
That is, high string frequency positively
impacted vowel reduction.
"Don't"
Phonetic reduction ranges from initial /do/
to /ɾo/ to /ɾə/ to /ə/ depending on the
frequency of the phrases in which it
appears. --Bybee and Scheibman (1999)
and Scheibman (2000)
cf. "I don't know/think/have (to)/want/like"
vs. "I don't need/follow/make contact"
Misunderstanding #6
That Word Frequency is independent of
Word Class in the lexical diffusion of a
sound change.
Modern Dutch diphthongization
of WGmc /î/
"We do not find a relationship between token
frequency and diphthongization....[H]ighly
frequent items such as krijgen, tijd (>200)
do not behave differently. . . from items
such as vijg (0) and vijl (2). And among
the items showing diphthongs rather often,
such as vrijdag, vrijen, vrij, fijn and vijf,
both low and high frequencies occur."
-- Goeman et al. (1993)
Adjectives
% Diphthongs
Adjectives (Celex frequencies)
66-77
vrij (9214), vrije (1406)
53-60
blij (4277)/blijde (225), fijn (2200)
43-51
grijs (197), rijk (1378), rijp (636),
rijpe (338), rijpst (0), stijf (1040),
wijd (1091)
Function words as a special case
Certain sound changes tend to affect
word categories with low sentence stress,
such as function words, but this does not
correlate necessarily with word frequency.
/an, on/ > /on, om/ in Pastoral Care (OE, late 9th c.)
(Phillips 2006)
By WORD
BY WORD
FREQUENCY
CLASS
Category
% <on, om> Freq
% <on, om>
Verbs
32
1-10
39
Nouns
44
11-20
49
Adjectives
44
21-60
53
Adverbs &
Function
words
84
61-400
80
over
400
98
/ø:/ > /e:/ in the Ormulum ( ca. 1180 a.d.)
(Phillips 1984; 2006)
BY WORD CLASS
BY WORD FREQUENCY
% <e>
Numerals
% <e>
0
1-10
62
Nouns
28
11-20
66
Verbs
67
21-30
45
Adjectives
70
31-60
43
Adverbs &
Function
words
100
61-90
28
91-400
21
Other Word Classes
Word frequency effects have been shown
to be clearest inside of word classes –
since in production word class is accessed
before the word form.
Specificially, the retrieval of a word's
syntactic information precedes that of
phonological information by approx.
40 msec. -- Turennout et al. (1998)
Misunderstanding #7
That analogy and borrowing suffice to
account for lexical diffusion (Kiparsky
1995; Janda and Joseph 2003).
Attribution of Lexical Diffusion effects to
Analogy & Borrowing
“[D]iffusionary effects in the spread of
phonological change through the lexicons
of speakers... are actually
epiphenomenal, being the result of
already-needed mechanisms of
analogical change and dialect
borrowing."
--Janda and Joseph (2003)
The Problem with Lexical Diffusion as
Analogical Change
Analogical changes affect the LEAST
FREQUENT words first--"where memory
fails"-- e.g., cows, dragons, aliens BUT
mice, geese, teeth;
Regular verbs (correspond, irrigate,
elucidate) vs. irregular verbs (be, go, see)
whereas, Lexical Diffusion often affects
the MOST FREQUENT words first.
Lexical Storage
Bybee, J. 2001. Usage-Based Phonology. p. 22:
Borrowing
Lexical diffusionists "see sound change . .
. as change affecting the sound in certain
words and then diffusing gradually to other
words in the lexicon. . . . This is like
'dialect borrowing', but with some words
borrowing from others in the same dialect.
It constitutes a different outlook on the
transition problem."
(Campbell 1998: 199)
But surely,
all change involves borrowing:
“[W]e have no criteria for determining
absolutely that there is an axiomatic
distinction between sound change and
borrowing (or contact change) because..
all changes must arise from contact
between speakers.” -- Milroy (1992: 88)
Wolfram and Schilling-Estes (2003):
a change becomes propagated
through social groups,
through phonological and
morphological environments, and
through the lexicon.
All of these happen simultaneously,
and even within individuals variation
is to be expected.
"[T]he notion of variability . . . applies to
both intra-speaker and inter-speaker
variation. In other words, an individual
speaker will go through a period of
fluctuation between the old and new
variant, and speakers within a given
speech community will show variation
from speaker to speaker with respect to
the use of the new and old variant" -Wolfram and Schilling-Estes 2003: 717.
As for changes that clearly imitate an
external dialect:
With regard to word frequency, there is
no difference between "borrowed"
changes and internal changes.
In fact, the spread of a sound change from
one dialect to another apparently follows
the same direction of diffusion as the
imitated dialect. E.g.,
Spread of broad /a:/ to the U.S.
fashionable import from Britain in the
eighteenth or nineteenth century.
A few words whose innovative
pronunciations were not borrowed (haft;
aster; rascal; vantage; ranch,
stanchion; Alexander, Flanders,
slander),
but only one (alas) that underwent the
shift in the Eastern U.S. but not in
England.-- Phillips 1989
Supports Trudgill’s (1986: 58) observation
that during accommodation people “modify
their pronunciation of particular words,...
with some words being affected before
others” , which implies that the spread of a
sound change from one group to another
must of necessity proceed via lexical
diffusion, affecting some words before
others.
Second point:
Of the subset of (already frequent)
words that had the innovative vowel
in British English, Eastern U.S.
English borrowed the most
frequent.
Eastern U.S. Words in
[æ]/[a:] for RP [a:]
Unchanged
Words with [æ]
Ave. Freq. of Words
with Borrowed [a:]
haft
Freq. of
Unchanged
Words with [æ]
0.00
aster
0.16
71.93 (-st)
rascal
1.22
29.65 (-sk)
vantage
0.75
36.71 (-nt)
stanchion
0.24
14.23 (-ntʃ)
Alexander; Flanders; slander
10.27; 1.47; 0.40
10.92 (-nd)
110.68 (-ft)
Therefore:
Since borrowing (i.e., imitation of other
speakers) is inherent to the spread of an
innovation from speaker to speaker
and since lexical diffusion is inherent in
the spread of an innovation from one
word to another
BOTH ARE INHERENT TO SOUND
CHANGE--not at odds with it.
AND:
Since the spread of all changes
depends on the similarities shared
by lexical entries,
Analogy, too, is endemic to sound
change.
Misunderstanding #8:
That age of acquisition or discourse
factors account for patterns of lexical
diffusion.
The AoA effect is reliable only for lowfrequency words
-- Gerhand and Berry (1999)
Gerhand & Berry (1998): naming latencies in the
reading by 33 British college students of 64
words, divided into the following 4 categories:
(a) early-acquired, high frequency (win, cousin);
(b) early-acquired, low-frequency (elf, rattle);
(c) late-acquired, high-frequency (sex, union);
(d) late-acquired, low frequency (cue, marvel)
They found that the two effects , , , were
"entirely additive: Participants were faster to
read aloud early-acquired than late-acquired
words and were also faster to name highfrequency than low-frequency words."
Evidence from Adult Vocabulary
Words for 'heaven', 'priest', and'devil'
pattern with high-frequency words in the
Ormulum ms., written by a monk.
Words that are frequent within a particular
occupation under reduction typical of
frequent words, as in boatswain /bosən/,
coxswain /kaksən/, and kiln /kɪl/.
Discourse Influence, as summarized by
Wright (2003):
"In these studies talkers have been shown
to produce more reduced speech when
contextual information within the
utterance or in the environment can aid
the listener in recognising what is said,
and to produce more careful speech
when the talker is aware of conditions
that may impede the listener's ability to
understand what is said."
Berg (1998: 243)
"Frequency speeds up the wordrecognition process . . . . If speakers
exploit this principle, they can be more
sloppy about the pronunciation of highfrequency words than about that of lowfrequency items, while still achieving the
same degree of communicative success.
In fact, this is precisely what speakers do:
the higher the redundancy of a word,
the less accurately it is articulated
(Lieberman 1963)."
Shortcoming:
Discourse considerations alone
do NOT account for other types of
changes, such as stress shifts or
vowel shifts.
Misunderstanding #9
That Frequency effects are found over a
population of speakers but not within
individual speakers.
“Labov’s (1994) assessment of the
‘neogrammarian controversy’ provides
striking confirmation of regular sound
change at the level of the individual,
and lexical diffusion at the level of
speech communities.” (Blevins 2004)
Fidelholz (1975):
Reduced vowel
Full vowel
astronomy
mistake
abstain
gastronomy
mistook
abstemious
Misunderstanding #10
that lexical diffusion is the diffusion of a
completed sound change (Blevins 2004)
OR
"all phonological change starts with lexical
diffusion and most ends up
Neogrammarian, given enough time"
(Lass1997)
Conclusions:
Phonological, morphological,
semantic, social, pragmatic, and
cognitive factors all influence which
words are affected when.
But Word Frequency must also be
recognized a key factor in how a sound
change spreads through the lexicon.
And . . .
The growing evidence of word
frequency effects in even very lowlevel phonetic output supports the
view that what seem to be
Neogrammarian/regular changes
must actually be changes that diffuse
very rapidly over words in the lexicon.
THE END