Threshold hypothesis and bottom-up processing in English listening of Japanese learners : High-frequency sound discrimination and categorical perception of consonants

在听力理解的加工模型中，熟练的学习者被认为具有自底向上的自动技能，这使他们能够有效地利用自顶向下的技能。也有人说，自底向上的技能必须达到一个阈值水平，自上而下的技能才能被使用。本研究主要关注听力“高频水平”的作用，并提出以下假设：(1)精通英语的日本听众对高频纯音有更好的辨别能力。(2)熟练的听者在单词水平上有更高的辨别辅音的能力。(3)高频纯音辨别能力与辅音辨别能力呈正相关。(4)熟练英语听者对高频纯音和辅音的辨别能力存在阈值水平。我们准备了2000Hz和3000Hz的纯声辨别测试和L-R分类测试，并分析了132名参与者的数据。在4种不同的速度下，用[光-右][游戏-祈祷]对L-R分类进行测试。此外，高/低频率声音的条件也受到控制。结果表明，GTEC测试结果表明，高熟练程度的被试(1)对高频纯音的辨别能力显著提高，(2)对辅音的辨别能力显著提高。对于假设(3)，相关性仅在2000 Hz声级上具有统计学意义，尽管它被认为不是很强。结果还表明，当低于1400 Hz的区域被切断时，只有高度熟练的听者才能区分出L-R音。这被认为支持假设(4)。

In the processing model of listening comprehension, proficient learners are considered to have automatic bottom-up skills, which enable them to utilize the top-down skills effectively. It is also said that the bottom-up skills must reach a threshold level for the top-down skills to be employed. The current investigation focused on the role of "high-frequency levels" of listening and posited the following hypotheses : (1) Proficient Japanese listeners of English have better discriminatory ability for high-frequency pure sounds. (2) Proficient listeners have higher ability in discriminating consonants at the word level. (3) Discriminatory ability for high-frequency pure sounds positively correlates with consonant discrimination. (4) There exist threshold levels for discriminatory ability for high-frequency pure sounds as well as consonant discrimination for proficient listeners of English. We prepared pure-sound discrimination tests for 2000Hz and 3000Hz and an L-R categorization test and analyzed 132 participants' data. L-R categorization was tested with [light-right] [play-pray] pairs at 4 different speeds. Additionally, conditions of higher/lower frequency sounds were controlled. The findings indicate that the high-proficiency participants, as measured by the GTEC : (1) discriminated high-frequency pure sounds significantly better and (2) showed significantly higher ability in discriminating consonants. As for the hypothesis (3), the correlation was statistically significant only at the 2000 Hz sound level, although it was not considered very strong. Results also indicated that only highly proficient listeners could discriminate L-R sounds when the areas lower than 1400 Hz were cut in the stimuli. This is considered to support Hypothesis (4).