How do ab-initio second language learners start to detect words? : an exploratory study on Russian

Abstract

The speech stream is a continuum and discrete units, e.g. words, cannot be identified from the signal alone. How language learners segment (i.e. recognise and store words) in the speech stream has typically been explored with respect to children (e.g., Jusczyk et al. 1994; 1999a,b). Researchers have only recently begun to examine how adult second language learners segment an unfamiliar natural language after ‘first exposure’ without instruction (Gullberg et al. 2010, 2012; Carroll 2012, 2013, 2014; Shoemaker & Rast 2013). I report on a study of how 28 English-speaking adults begin to segment words after hearing them in fluent Russian speech during four sessions. The study explored the following questions: (1) Does participant' ability to identify words increase over sessions? (2) Do participants rely on segmentation cues such as phonotactics, word-initial stress, and word length? (3) If so, how do these cues interact? (4) Can learners generalise to the novel examples? (5) Are there differences between linguistically trained and naïve participants? Each day for four successive days, 28 participants were exposed to audio input in Russian for seven minutes (= 28 minutes exposure). Input comprised of 48 sentences of natural speech with target words embedded in a sentence medial position. After each exposure phase, participants were tested on their detection abilities of words they heard in the input as opposed to words they did not hear using three tasks: a word recognition task, a forced-choice task, and a cognate identification task. The word identification and the forced-choice tasks investigated if participants could detect words they heard in the input as opposed to words they had not heard. The purpose of the cognate identification task was to eliminate those participants who might not have been paying sufficient attention to the input (which was uncontrolled in the previous studies on first exposure). A word recognition and a forced-choice task conducted each day showed that segmentation improved significantly over time. Segmentation patterns reflected the influence of English phonotactics, sensitivity to weak-strong stress, and the interaction of the two, which, particularly for the word recognition task, stems from participants subconscious analysis of Russian. Also, participants could generalise phonotactic patterns of Russian to novel words. The study did not find a difference between linguistically trained and naive participants. The study concludes that beyond native language bias, adults deploy the various segmentation mechanisms similar to those children use.