Copyright 2000 Psychonomic Society, Inc. 834
Perception & Psychophysics
2000, 62 (4), 834-842
The acoustic/phonetic characteristics of speech vary as
a function of speaker, rate of speech, prosody, and so forth.
Yet, when we process our native language, we are hardly
ever aware of such variability; indeed, these variations
are apparently dealt with automatically and effortlessly
by the perceptual system. However, when processing ar-
tificially degraded speech or when listening to speakers
with foreign accents, it is more difficult to make suitable
adjustments. Schwab, Nusbaum, and Pisoni (1985) suggest
that several sentences are required to adjust to syntheti-
cally generated speech. Anecdotally, listening to speech
spoken with a foreign accent can also take some time be-
fore it becomes fully intelligible. What are the mech-
anisms responsible for such slow adjustments? Why are
some adjustments easier than others? For instance, for a
native speaker of English, English spoken with a Dutch
accent seems far easier to understand than English spo-
ken with a Japanese accent. Why?
Several studies have shown that language representa-
tions in adults are, to some extent, language specific. Lis-
teners behave as if they process speech sounds through
the filter of phonemic categories of their maternal lan-
guage and have difficulty processing nonnative contrasts
(Best, McRoberts, & Sithole, 1988; Goto, 1971; Mann,
1986; Miyawaki et al., 1975; Pallier, Bosch, & Sebastián,
1997; Sebastián-Gallés & Soto-Faraco, 1999; Werker &
Tees, 1984). Similar filters operate at the syllabic and
suprasegmental levels of representation (Cutler, Mehler,
Norris, & Seguí, 1983, 1989; Dupoux, Christophe, Sebas-
tián, & Mehler, 1997; Otake, Hatano, Cutler, & Mehler,
1993; Pallier, Sebastián, Felguera, Christophe, & Mehler,
1993). Moreover, lexical access strategies are also lan-
guage dependent. That is, listeners exploit language-
specific cues to find word boundaries in the continuous
signal (Cutler & Norris, 1988; McQueen, Norris, & Cut-
ler, 1994; Norris, McQueen, & Cutler, 1995).
Cutler & Mehler (1993), following the proposals of
Abercrombie (1967) and Pike (1946), have assumed that
all of the above differences between languages can be ac-
counted for by positing the existence of a small number
of broad classes of languages, defined rhythmically. Spe-
This research was supported by grants from the Human Frontier Sci-
ence Program and the Spanish Ministerio de Educación y Ciencia (Con-
tract PB97-0997) and the Catalan Government (Grup de Recerca Con-
solidat 5120-UB-05). We thank T. Otake (Dokkyo University), K. Forster,
and M. Garrett (both at the University of Arizona) for their help in
preparing and recording the Japanese and English materials. A.C. is cur-
rently at the Psychology Department, Harvard University. Correspon-
dence concerning this article should be addressed to N. Sebastián-Gallés,
Universitat de Barcelona, P. de la Vall d’Hebron 171, 08035 Barcelona,
Spain (e-mail: sebastia@psico.psi.ub.es).
Adaptation to time-compressed speech:
Phonological determinants
NÚRIA SEBASTIÁN-GALLÉS
Universitat de Barcelona, Barcelona, Spain
EMMANUEL DUPOUX
Laboratoire de Sciences Cognitives et Psycholinguistique, EHESS–CNRS, Paris, France
ALBERT COSTA
Universitat de Barcelona, Barcelona, Spain
and
JACQUES MEHLER
Laboratoire de Sciences Cognitives et Psycholinguistique, EHESS–CNRS, Paris, France
Perceptual adaptation to time-compressed speech was analyzed in two experiments. Previous re-
search has suggested that this adaptation phenomenon is language specific and takes place at the
phonological level. Moreover, it has been proposed that adaptation should only be observed for lan-
guages that are rhythmically similar. This assumption was explored by studying adaptation to differ-
ent time-compressed languages in Spanish speakers. In Experiment 1, the performances of Spanish-
speaking subjects who adapted to Spanish, Italian, French, English, and Japanese were compared. In
Experiment 2, subjects from the same population were tested with Greek sentences compressed to two
different rates. The results showed adaptation for Spanish, Italian, and Greek and no adaptation for En-
glish and Japanese, with French being an intermediate case. To account for the data, we propose that
variables other than just the rhythmic properties of the languages, such as the vowel system and/or the
lexical stress pattern, must be considered. The Greek data also support the view that phonological,
rather than lexical, information is a determining factor in adaptation to compressed speech.

PHONOLOGICAL ADAPTATION TO TIME-COMPRESSED SPEECH 835
cifically, three classes of languages have been proposed,
each of which triggers the use of dedicated processing
mechanisms. Languages such as French, Spanish, Cata-
lan, Italian, and Portuguese would fall into the syllable-
timed class; languages such as English and Dutch would
fall into the stress-timed class; and languages such as
Japanese would fall into the mora-timed class. If such a
claim is correct, one expects to find that processing a lan-
guage that belongs to the same class as the maternal
language should be easier than processing more distant
languages. If so, one could use the perception of degraded
signals across languages to explore in more detail the lan-
guage class hypothesis and, more generally, the influ-
ence of rhythm/phonology on processing.
Adaptation to time-compressed speech is a useful
method by which to address these issues. In these exper-
iments, participants listen to compressed sentences and
try to understand what is being said. A previous study
(Mehler et al., 1993) showed that, in the context of high
presentation rates, Spanish, Catalan, French, and English
subjects adapt their perceptual system and increase their
recall performance after listening to a few sentences. In
this study, French and English monolingual subjects were
presented with sentences compressed at different rates
(uncompressed, 50%, and 40% of their original length)
and then were tested on sentences compressed to 40%.
Savings were observed for the groups with 50% and 40%
compressed sentences, but not for the group with the un-
compressed ones, as compared with the performance of
a group of subjects who had not been previously exposed
to any kind of adaptation materials. Analogous results
were obtained for Spanish–Catalan bilinguals. Interest-
ingly, these subjects also benefited from being exposed to
compressed materials in their second language (mono-
lingual Spanish subjects also benefited from being ex-
posed to compressed Catalan, a language that they did not
understand). These results were in contrast to the lack of
transfer observed with both monolingual and bilingual
French and English subjects, when they were exposed to
compressed English or compressed French, before being
tested in their maternal languages. Although there were
some problems with this study (the experimental proce-
dures used were not the same in all of the different ex-
perimental conditions, thus making it difficult to com-
pare the performance of the different groups of subjects
being tested), the original results have been confirmed in
other studies.
Dupoux and Green (1997) expanded the above obser-
vation and found that as few as 5–10 sentences are suf-
ficient for adaptation to take place. Moreover, they no-
ticed that neither a change of speaker nor a change in the
compression rate significantly affects adaptation. Du-
poux and Green argued that these effects show that adap-
tation takes place at a relatively abstract level. In a study
by Altmann and Young (1993), English speakers tested
with compressed materials made from nonsense words
showed savings from exposure to these compressed ma-
terials; therefore, it seems that lexical access and/or com-
prehension does not seem necessary for adaptation to
occur. Pallier, Sebastián-Gallés, Dupoux, Christophe,
and Mehler (1998) also replicated and extended some of
the previous findings of Mehler et al. (1993). Interest-
ingly, they observed benefits when monolingual British
subjects were exposed to highly compressed Dutch be-
fore being tested with English sentences.
So far, all of these experiments have shown that sub-
jects can adapt to fast speech even with signals that do
not belong to their mother tongue. Could it be that adap-
tation to compressed speech is a general phenomenon that
applies for all speeded signals, irrespective of rhythm/
phonology?
Two studies suggest that this is not the case. Aside
from the original study of Mehler et al. (1993), Pallier
et al. (1998) also found a case in which compressed speech
in one language failed to trigger adaptation in another lan-
guage—namely, French and English showed no cross-
adaptation. These results suggest that, first, adaptation
to speeded signals is not a general phenomenon but, rather,
depends on the linguistic properties of the adapting signal
and second, that the relevant properties are rhythmical/
phonological: English and Dutch are in the same rhyth-
mic class, as are Catalan and Spanish, and both show
adaptation. English and French are not in the same rhyth-
mic class and show no adaptation. However, alternative
accounts are conceivable, and the English–French study
has several limitations.
First, even though comparable compression rates were
used for English and French in the Pallier et al. (1998)
study, English globally yielded very poor performance,
as compared with French. This effect could have resulted
from the fact that English is intrinsically less compressible
than French or, more plausibly, from some initial differ-
ences in the uncompressed French and English materials.
In any event, this difference in base performance raises
the possibility that the compressed English materials
may have been too degraded to function as adaptation ma-
terial. It is, then, important to replicate this finding with
adaptation materials that are still perceptible.
Second, the absence of English–French cross-adaptation
could be related to the fact that, in this particular exper-
iment, the subjects were bilinguals. Bilinguals may use
one processing strategy for one language and apply it—
inadequately—to the other. This would yield a cost that
could potentially cancel out any purely phonological
benefit from the adapting sentences. Such a processing
cost would not arise in monolinguals, who would not even
attempt to process the foreign language beyond the phono-
logical level. True enough, Pallier et al. (1998) tested
Catalan–Spanish bilinguals and found, in this population,
cross-adaptation between Catalan and Spanish stimuli.
But French and English have less overlap in morphology
and lexical roots than do Catalan and Spanish, and this
difference might play a role in bilingual processing (more
on this problem below). Moreover, all of the published
monolingual studies have found an adaptation effect across
languages. Thus, it is important to replicate differences