Impact of computer-assisted pronunciation training on child English language learners

Abstract

Understanding second language speech has been a pressing issue for researchers. Accounts for sources of error shown by L2 learners include age of initial exposure, relative markedness, L1 functional constraints, speci cally perceptual salience and frequency (Colantoni and Steele, 2008), and perception, which is the basis for explaining cross-linguistic in uence by most L2 speech learning theories (Colantoni, Steele, and Escudero, 2015), but they do not include the delay of oral production. When we look at younger beginner L2 learners, L1 in uence can also be observed. The aim of the present study was to address the impact of delaying oral production and for this computer-assisted pronunciation training (CAPT) was used on Arabic-speaking children in Libya learning English as a second language with no prior instruction in English. English instruction in Libya is typically delivered by non-native teachers whose non-native input is also a possible source of L1 in uence. The software provided native speaker input to address this additional aim. Within the software, test words were presented in orthographic and audio formats with pictures depicting meaning. Predictions on the role of output have varying underlying assumptions. Proponents of the importance of production practice such as Swain (1985; 1995; 2005) and Mackey (2007) argue that it is a tool for creating novel linguistic knowledge and promoting cognitive processes (see Colantoni and Steele, 2008 in their Hybrid model). Following a three-week training with use of the software, 38 seven-year-old participants took part in picture-naming, read aloud and delayed repetition tasks in an immediate post-test and, of these, 30 took part in similar tasks for a delayed post-test 10 weeks later. The 38 participants were divided into two training conditions, Listen and Speak (n.=20) and Listen Only (n.=18) to test the role of delayed production on L2 learning. Another group of 20 aged-matched participants took part in a three-week training with use of Traditional Teaching and participated in the same tasks in an immediate post-test and of these 18 took part in similar tasks for a delayed post-test 10 weeks later. The Traditional Teaching condition was added to compare input type on participants within the same age group. The aspects of pronunciation measured were target-likeness rating, match rating, various acoustic cues including Voice Onset Time (VOT), vowel-onset fundamental frequency, and spectral tilt (Ahi-A23). The participants' L2 values were compared to the target language and their L1 values to test predictions made by models of speech learning. The phonetic data revealed signs of merger categories between L1 and L2 corroborating the ndings of Flege (1995) and MacKay, Flege, Piske, and Schirru (2001). Additionally, phonological processes were examined and compared to processes found in L1 English child phonology. The amount of lexical learning was also explored. Results for TL-likeness and match rating revealed that the experimental conditions statistically outperformed the Traditional condition in both tests. In the delayed test however, the Listen and Speak condition statistically outperformed Listen Only participants, who continued to outperform the Traditional learners. VOT and vowel-onset f0 analyses revealed that participants from all training conditions failed to establish independent L2 categories. Rather, they illustrated intermediate values resembling both native and target phonetic categories. In terms of lexical learning, the experimental conditions outperformed the Traditional condition in terms of the amount of fully learned words in the delayed repetition and picture-naming task but they all performed the same in the read aloud task. Some interlanguage processes were demonstrated by the learners in addition to the expected transfer from their Arabic variety. These varied depending on the sound class and conformed to universal language development and input from native speakers of the target language. It is concluded that the ndings support the importance of output in language learning for L1 beginning-level children in the classroom as suggested by the Hybrid model (Colantoni and Steele, 2008).