Paul & Paola's second web site

One-to-multiple vowel mapping in the perception of Dutch learners of Spanish

Paola Escudero, escuderopaola@hotmail.com
Paul Boersma, paul.boersma@hum.uva.nl

September 13, 2001, 7.20 GMT

Note: This web page contains all the material from Paul & Paola's experiments,
but is in some respects preliminary.
Please do not cite. You can cite the available papers instead.

1. Abstract
2. Recordings
3. Production analysis
4. Listening experiment
5. Analysis of category use
References

1. Abstract

It has been widely suggested that learners rely on the perception strategies of their first language when learning to perceive second-language sounds and that they will be succesful perceivers/lexicalizers if they can map each L2 phoneme on a different phoneme in their L1. Nevertheless, we will show that L2 perception can be problematic if an L2 phoneme has multiple correspondents in the native language. As an example, we tested the perception of Spanish vowels by Dutch learners of Spanish (in particular, the mapping of Spanish /i-e/ on Dutch /i-I-E/).

Forty Dutch learners of Spanish were tested in three listening tasks, each of which contained 160 CVC stimuli in random order. Of these, 80 were taken from naturally produced Spanish /i/, /e/, /o/, /u/ in CVC contexts. The remaining 80 stimuli were either Dutch-sounding or Spanish-sounding fillers, depending on whether we wanted the listeners to enter a Dutch or a Spanish perception mode.

The first and second tasks aimed at assessing the difference between the learners' Dutch and Spanish modes of perception. In the first task, the listeners were presented with the 80 target stimuli mixed with 80 Dutch-sounding fillers, embedded in a Dutch carrier sentence, and they were asked to orthographically label the Spanish tokens, choosing from the 12 Dutch vowel symbols. In the second task, the listeners were presented with the same 80 target stimuli, but embedded in a Spanish carrier sentence and mixed with 80 Spanish-sounding fillers, although they still had to use the Dutch response categories. It turned out that most of the subjects had different perception strategies according to the language that they thought they heard: not only did most learners use fewer categories in the second task than in the first, but many also shifted their perceptual category boundaries between tasks.

The third task aimed at assessing the learners' L2 identification performance. The learners were presented with exactly the same stimuli as in the second task, but this time they were required to respond with Spanish orthography. The fraction of correct responses turned out to correlate with the size of the difference between the first and second tasks, i.e. with the difference between the Dutch and Spanish perception modes. Several learners had a markedly poorer identification for front vowels than for back vowels. A comparison with the second task shows that this effect correlates with the learner's use of the middle category /I/, on which both Spanish /i/ and /e/ are often mapped. We conclude that the availability of extra categories can render the advantageous one-to-one mapping strategy (i.e, Spanish /i-e/ to Dutch /i-E/) inaccessible to the learner.

2. Recordings

We recorded the Spanish text estimadosBabeles.txt and the Dutch text voedselrelletjes.txt. These texts were modified from their originals, so as to contain several instances of the phrases "luister naar" and "la palabra", which could be used as carrier phrases later on. The bilingual speakers was CLL, native from Spain, lived in the Netherlands from the age of 12, currently a university teacher of Spanish. Both texts were recorded twice, and the carrier sentences "luister naar..." and "la palabra..." were also recorded separately several times:

From the recordings, we selected (17.2 MB) as the best Spanish, and recordings/cll_voe2.wav (9.7 MB) as the best Dutch. We chose the following instances as carrier phrases: speakers/cll/cll_luisternaar.wav (117 kB) and speakers/cll/cll_lapalabra.wav (84 kB).

3. Production analysis

We segmented most CVC chunks in the Spanish text with the help of the Praat program. The TextGrid had three tiers, because the chunks tended to overlap (both consonants were intended to be audible). The three CVC tiers contained labels of the form C₁VC₂, where V was any of the five vowels {a,e,i,o,u}, and C₁ and C₂ were usually transcribed with a single ASCII symbol:

T, x: the voiceless dental and uvular fricatives.
B, D, G: the voiced bilabial, dental, and velar approximants.
ch: the voiceless post-alveolar affricate.
j: the palatal glide.
r: the tap.
rr: the trill.

After each of the three CVC tiers, we inserted a point tier that marked the centre of the vowel in each interval in the tier above it. The point was labelled with a "Spanishness" judgment by the Dutch-speaking author, which designated the degree to which the consonants could pass for Dutch vowels:

1: very Dutch and very un-Spanish.
2: rather un-Spanish.
3: could be Dutch as well as Spanish.
4: rather un-Dutch.
5: very un-Dutch.
6: the Dutch consonant category is useless for our purposes, e.g. if Spanish [B] or [D] sound like Dutch /w/, /l/, or /j/.

Some CVC chunks had a double score, e.g. "46". In the Spanish text, the scores were chosen from "3", "4", "5", "6", "46", and "56". Ideally, we wanted to use only the ones marked "3" for our target words, because these were the ones that could pass for Dutch in the first, for Spanish in the second experiment.

We performed an analogous segmentation of the Dutch text, choosing scores from "1", "2", "3", "16", and "26".

Here are the resulting TextGrids:

The Praat script getProductionData.praat extracts all the intervals as WAV files into the directory speakers/cll/est2 or speakers/cll/voe2, and creates a table with stimulus name (= file name), first formant (F1), second formant (F2), and Spanishness, e.g. lines like:

"cll/est2/u008.13_kus" 331 1212 3

This means that the file u008.13_kus.wav (the chunk [kus] with the vowel /u/, starting at 8.13 seconds into cll_est2.wav) has a vowel with an F1 of 331 Hz, an F2 of 1212 Hz, and a Spanishness judgment of 3. The tables are:

4. Listening experiment

The stimuli are in stimuli_winsit.exe, which is a self-extracting archive. Download this e.g. to your desktop and unpack it by double-clicking. You will see a new folder called "stimuli" on your desktop. It contains 182 WAV files.

The listening experiment, which can be run in Praat, is bu.ExperimentMFC. This can be read with "Read from file...", then run. In this experiment file, you will see that the stimulus folder is set to "C:\Windows\Desktop\stimuli\". You can change this if your desktop is somewhere else or if you saved the "stimuli" folder somewhere else, e.g. "D:\p2b\stimuli\" if you saved the "stimuli" folder in the folder "p2b" on disk "D:".

After running, the experimenter extracts the three results and saves them together in a Collection object with "Write to short text file...". Here are the results of a pilot experiment run in June 2001:

5. Analysis of category use

The script measureEnthropies.praat measured for each listener the number of categories that s/he used in the first and second experiment:

Responses Enthropy1 Enthropy2 Categories1 Categories2
1 fm        2.76       2.67       6.8       6.3
2 jdg       3.14       2.75       8.8       6.7
3 dw        3.08       3.13       8.5       8.8
4 ds        2.91       2.74       7.5       6.7
5 mvb       2.71       2.57       6.5       5.9
6 evd       3.24       3.07       9.4       8.4
7 mb        2.79       2.65       6.9       6.3
9 is        3.06       2.93       8.3       7.6
10 mk       2.95       2.93       7.7       7.6
11 ben      2.98       2.86       7.9       7.2

The correct results for the second part of the experiment are correct.ResultsMFC. These have an enthropy of 2.16, i.e. 4.5 categories.

We see that the listeners use fewer categories when they think that the language is Spanish than when they think that the language is Dutch, although they were asked in both cases to judge the nearest Dutch category. This effect is not due to fatigue or a learnign effect, since it did not occur with those listeners who did not know Spanish (i.e. listener dw, who did not know what "la palabra" meant).

References

Bradlow, Ann (1995): A comparative study of English and Spanish vowels. Journal of the Acoustical Society of America 97: 1916-1924.

Bradlow, Ann (1996): A perceptual comparison of the |i|-|e| and |u|-|o| contrasts in English and in Spanish: Universal and language-specific aspects. Phonetica 53: 55-85.

Cenoz, J. & L.G. Lecumberri (1999): The effect of training on the discrimination of English vowels. International Review of Applied Linguistics in Language Teaching 37: 261-275.

Flege, James E., M.J. Munro & R.A. Fox (1994): Auditory and categorical effects on cross-language vowel perception. Journal of the Acoustical Society of America 95: 3623-3641.

Johnson, Keith (2000): Adaptive dispersion in vowel perception. Phonetica 57: 181-188.