Turning Dr. Strangelove into George W. Bush: Determinants of accentedness
Tuesday, June 02, 2015
There are some 40,000,000 foreign-born people living in the United States today. Most of those people hail from Asia or Latin America and the Caribbean, and their native language is therefore most likely not English. And indeed, it is not uncommon for people in the U.S.—and also the U.K., Canada, and Australia—to speak English with a foreign accent.
Speaking with an accent is not inconsequential, as Cassie Jacobs reported in a recent post here:
“One consequence of sounding like a non-native speaker is that we are trusted less, and what we say is not taken at face value. One particularly striking example is that foreign accents make us less likely to believe statements made by non-native speakers, as measured by a truth rating task. For example, if you hear a statement that ‘Ants don't sleep’, you will likely rate the truth value of that statement lower if it is spoken with a heavy foreign accent than without. This is true even when we are merely summarizing facts said by native speakers.”
So what determines one’s accent? What features of one’s native language (usually called “L1”, for “Language 1”, in the literature) contribute to one’s accent in “L2” (i.e., the second language, namely English in most studies)?
There are some intriguing and sometimes counter-intuitive aspects to this question. For example, speech sounds that are similar between the two languages may interfere with each other because learners may consider the L2 version to be a realization of the comparable sound in L1, even though the two differ in subtle ways. Specifically, the French sound /u/, as in “tous”, at first glance appears to be the same as the English /u/, as in “too.” However, in actual fact those two variants of /u/ differ in the way in which they are pronounced by native speakers of French and English. In consequence, when native English speakers pronounce “tous” they make it sound like “too” and hence produce an accented utterance.
Intriguingly, this does not mean that native English speakers always fail to “sound French”: When they produce the sound /y/, as in “tu”, which has no equivalent in English, native English speakers can produce this new and unfamiliar sound with greater accuracy than the /u/. Give it a try; you may soon sound like the native French speaker in this video.
Recent research published in the Psychonomic Society’s journal Attention, Perception, & Psychophysics sheds further light on the factors that determine the accentedness of L2 speech. Researcher Vincent Porretta and colleagues differentiated between two classes of factors: Those that were tied to the characteristics of the speaker—and hence affected the acoustics of their speech production—and those tied to the features of the language itself—factors such as the frequency with which words occur in the language.
Porretta and colleagues used a very simple experimental task: Participants listened to recordings of words spoken in English by 9 different native Chinese speakers, and rated their accentedness on a scale from 1 (no accent) to 9 (strong foreign accent). For control purposes, a native English speaker was also included in the set of stimuli.
The focus of the research was on measuring a large number of acoustic variables and language attributes to predict the participants’ judgments of accentedness.
The results showed that both acoustic and lexical variables affected the perception of gradient foreign accentedness. While this may appear unsurprising at first glance, it is actually quite an intriguing finding.
To understand why, we must realize that judgments of accentedness require that the speech signal (i.e., the single spoken word in this instance) is matched to the receiver’s internal representation of what constitutes a native-like production. The closer the match, the less strong the accent.
It is easy to see why the acoustic properties of the signal—for example, whether a talker says “too” rather than “tous”—determine the ease of matching. It is, however, somewhat more surprising that the ease of matching should also be determined by the linguistic and lexical properties of the speech. After all, those factors are entirely independent of the speaker’s ability to phonetically approximate native productions.
To illustrate, one of the variables examined by Porretta and colleagues was “phonotactic probability.” Phonotactic probability refers to the frequency with which a sound, and a sequence of sounds, occur in a given position in a word. For example, “sit” has a high phonotactic probability because many other words have those sounds in the same position (i.e., initial /s/, final /t/, and so on), whereas “these” has a low probability because it contains sounds that infrequently occur in other words in the same position and that infrequently co-occur.
Porretta and colleagues found that when the phonemic sequence was more probable, perceived foreign accentedness was reduced. They interpreted this as a “frequency-like effect that facilitates the matching process and the judgment of wellformedness with respect to nativeness.” That is, a person whose native language is something other than English will be perceived as having less of an accent if they pronounce words whose statistical structure is very common—and crucially, this advantage was independent of the speaker’s actual ability to reproduce native sounds. Thus, the matching of a speech signal to internal representations also depends on variables that have nothing to do with the speaker’s abilities.
Turning to acoustic variables related to a speaker’s ability, one of the important variables involved the duration of words and vowels. Specifically, the proportion of a word that was subsumed by the duration of the vowel, and how distant this proportion was from the typical native value, was an important determinant of accentedness. This is again somewhat surprising because duration of a vowel is independent of its sound: It appears that speakers who are more able to approximate a native durational pattern are perceived as being less accented, independent of how close in actual frequencies their “too” comes to “tous” (or vice versa).
What are the implications of the work by Porretta and colleagues on people who struggle with their foreign accents?
Two potential solutions come to mind: First, irrespective of how difficult it is to say the vowels in “tous” or “tu”, get the timing right. That may be half the battle. Second, when you can choose your words, choose a vocabulary that’s statistically likely.
Focusing on that may ensure you a place on the list of actors who can hide their accents perfectly.