|Discussion Topic 2:
Overall degree of L2 foreign accent
Degree of L2 foreign accent
Above: the main square
in Tuscania; the Post
We carried out quite a few studies in which overall degree of perceived foreign accent
(henceforth, FA) was scaled. In some studies the main purpose of the study was to scale
FA, which was intended to serve as a measure of overall degree of L2 speech production
accuracy. In other study, degree of FA was assessed for other reasons, for example, to
assign prospective participants (Ss) to relatively proficient and non-proficient subgroups.
The first step was to record speech materials that had been produced by nonnative Ss
and a control group of native speakers of English. The non-natives all spoke Italian as an
L1. They were selected on the basis of their age at the time they immigrated to Canada
(AOA, age of arrival, 10 subgroups of n = 24). All nonnative Ss responded to an LBQ
(language background questionnaire). The speech materials were rated by native English
speakers residing in the same town in Canada where the nonnative Ss had lived for
decades. The primary aim of the study was to determine which was the first AOA-defined
subgroup to receive significantly lower ratings than the native speakers. Another aim was
to determine if degree of FA increased notably at some point along the AOA continuum,
for example, at 12 years.
Great care must be taken not only in deciding what speech materials to record but also
how to elicit them. The key is to assure a level playing field for all participants, both native
and nonnative. For example, if a difference between Ss are likely to exist in reading ability,
written materials should NOT be used to elicit L2 production.
This blog post provides a summary of factors that have been identified in the literature as contributing to
variation in degree of FA. Grosjean's blog is an excellent source of readable information on many aspects of
bilingualism. It goes well beyond the narrow focus of this site, namely, phonetic aspects of bilingualism. [JEF
Jan. 20, 2013]
Factors affecting degree of foreign accent in an L2
Some researchers believe that acoustic measurements are the only truly "objective" measure of L2 production.
For them, listener judgments are "subjective" and thus less useful and perhaps not even valid. I don't agree.
Just as care must be taken in making acoustic measurements (e.g., choosing the correct number of LPC
coefficients when measuring vowels), so too care must be taken when eliciting and analyzing listener judgments.
Flege et al. (1995) used a delayed repetition task to elicit the production of English native
speaker of English ask In which direction did he turn? followed by a female voice He
turned to the rightFlege et al. (1995) used a delayed repetition task to elicit the
production of English . After hearing the He turned to the right. After hearing the (male)
question a second time, the Ss were to say aloud the answer furnished seconds earlier by
the female voice.(male) question a second time, the Ss were to
The elicitation task, therefore, can be characterized as a delayed sentence repetition
task. We reasoned that furnishing a model of the English sentences to be produced (the
female voice) would greatly reduce intonational as well as temporal variability. Atlhough
intonation and rhythm are both important and represents important topics for study, we
wanted to avoid differences of this kind as well as substantial differences in sentence
duration. All of these might have provided a cue to FA apart from segmental production
accuracy, the focus of the study.
Flege et al. (1995) reasoned that the presence of a filled interval (the male voice) that occurred between the
model and its repetition would prevent the Ss from mimicking the model from sensory memory without
reference to their own internalized sensory motor representations of L2 speech.
Some general considerations
To facilitate the task, the five sentences examined were all simple and short. The sentences were elicited
multiple times. The second (or, if necessary, the third) repetitions were used in analyses assuring that all
sentences were produced fluently, with no missing data.
The test sentences were not constructed in such a way as to maximize production difficulty by containing a
large number of segments known to be difficult for native Italian learners of English. (Such a strategy might be
opportune, for example, in a study of early learners likely to have a relatively good pronunciation of the L2).
However, the phonetic material present was sufficient to allow the native English listeners to differentiate
accented from non-accented sentences.
Flege et al. noted relatively little difference across the five test sentences examined, but larger and more
important differences between the 10 native English-speaking listeners who rated the sentences. All 10
listeners gave progressively lower ratings -- indicating increasingly strong FA -- as the AOA (age of arrival) of
the Italians increased. However, some listeners seemed to be more senstive to variation in FA than others.
Accordingly, Flege et al. (1995) examined the ratings of the 10 listeners separately in order to address their
research question, namely, what was earliest age of L2 learning at which FAs are likely to be heard? We used
Williams' test to determine which of the 10 native Italian (NI) subgroups defined by AOA was the first to differ
from the native English(NE) control group (see Figure 3). One listener distinguished all 10 groups of native
Italian Ss, including those who had arrived in Canada as young children and had been immersed in English for
decades. For three other listeners, the AOA of the first foreign-accented NI group was 5.2 years. The least
sensitive listener gave lower ratings to the NI Ss who arrived at 11.6 years and subsequently.
The answer to the "When do FAs become inevitable?" question clearly depended on the listener. If we had
accidentally selected just the three most sensitive listeners, we would have concluded that FA first becomes
evident at am average age pf 4.5 years. Had we selected just the three least sensitive listeners to form the
"panel", our estimate would have been 10.3 years.
Over the years, many colleagues and not a few reviewers have asked me why we didn't simply measure the
sentences acoustically for FA. My answer is that there would not have been any "simple" about it. Let's assume
that six of the 13 phonetic segments in the test sentence He turned to the right would manifest native vs.
nonnative differences in a detailed acoustic analysis. How great would an acoustic divergence from the
phonetic norms of English need to be in order to "count" as a FA marker? How many such markers would be
needed to classify a sentence as foreign accented? All? Half? Perhaps more importantly, which markers did the
listeners actually use when rating the sentences?
difficulty generating usable findings. Acoustic analysis might in some cases take precedence when the focus
are segment sized units of L2 speech. For larger units, however, listener judgments provide the gold standard
in my opinion. For an elaboration of this topic, see the detailed analysis of the word taco as produced in both
Spanish and English by Spanish-English bilinguals (Flege & Munro, 1994.)
Are phoneticians clueless?
On July 15, 2015, Leonhard Klaar (firstname.lastname@example.org) posted comments under the title "The origin of
accent" Mr. Klaar's post ended with the provocative question: Why do phoneticians not want to admit that they
have no idea about the origin of accent?
English word "two" or how native speakers of Spanish pronounce the word "taco".
From your comments, it seems that you consider the primary cause of foreign accent to be an-related loss of
ability to learn to pronounce sounds differently than in the L1, that is to learn to articulate certain L2 sounds in
a native-like way. New wine in old bottles.
The conclusion I drew from nearly 30 years of research, however, was that FA is not due to difficulty in learning
to pronounce sounds after a "critical period". One major cause of adult-child differences in L2 learning is that
adults, and they are more likely than adults to hear L2 sounds pronounced correctly by native speakers (as
opposed to hearing incorrect productions from other nonnative speakers).
Another important cause of FA is that as children's L1 develops, their phonetic categories develop. As this
happens, it becomes more likely that L2 sounds that happen to be unlike any found in the L1 will be produced
using a sound from the L1 inventory. It's not always "new wine in old bottle" for adults, but the range of L2
sounds that are fully learnable seems to diminish as the L1 phonetic system develops. In sum, children are
more likely to develop correct perceptual models for L2 sounds than adults do because they receive more and
better input and because their existing perceptual representations for L1 sounds are not yet fully developed.
When and if correct perceptual models (phonetic categories) are established, adequate articulatory motor
program can be developed.
I was surprised to read your comment on the posterior pharygeal wall. It's not something people think about a
lot. However, it is quite relevant. In 1987 we published a detailed study examining four groups, native speakers
of English (adults, children) and native speaker of Chinese (adults, children). Our study focused on production
of the consonant /b/ at end of words like “bob”. Producing /b/ in word-final position is far more difficult than one
might imagine. To keep the vocal folds vibrating when the lips are sealed, it is necessary to maintain higher
pressure below than above the level of the vocal folds.
The English adults had learned how to do this by actively expanding the pharyngeal cavity at exactly the right
moment in time. As a result, they managed to sustain closure voicing during labial closure and so their /b/s
were heard as /b/, not as the voiceless consonant /p/. The English children were learning this little trick too, but
had not yet fully mastered it. The Chinese children were behind the English children in this aspect of speech
motor control, but slightly ahead of the Chinese adults. We attributed the observed differences in articulation to
how much English language input members of the four groups had received in their lifetimes.