The Influence Of Chinese Semantics On Early Automatic Speech Recognition

Phoneme merger refers to the phenomenon that two phonemes are pronounced distinctly different in one language or dialect,but difficult to distinguish or merge into one phoneme in another language or dialect.This interesting linguistic phenomenon often occurs not only in the learning transformation of two different languages,but also in two different dialects of the same language.Why is it so easy for some people to distinguish between phonemes that are difficult for others to distinguish? What happens in the brain of the phoneme merged speakers? Early studies on phoneme merging mainly focused on sociolinguistics.Psychological studies mostly stopped at the psychophysical level.Later,a small number of researchers began to use electrophysiological techniques to explore this issue,but the number of studies was small and the results were controversial.So up to now,there is no clear explanation for the neurological mechanism of phoneme merger.In this study,the common phenomenon of /n/-/l/ phoneme merger(/n/-/l/ indiscriminate)in Chinese dialects was selected as the research object,and the ability of phoneme merged speakers and unmerged speakers to distinguish and process merged phonemes at the phonological and semantic levels were compared.In experiment 1,we explored the cognitive processing characteristics and neurological activities of /n/ and /l/ phonemes in the pre-conscious stage of the 15 merged speakers(6 males) and 15 unmerged speakers(4 males),and compared whether there were differences in EEG responses between the two groups.Using the passive oddball paradigm and the mismatch negative(MMN)waves as the main electrophysiological index,the participants' automatic processing of Chinese pronunciation /liú/ and /niú/ was recorded.EEG results showed that unmerged speakers evoked significant MMN at 170-220 ms after stimulation,while merged speakers did not exhibit significant MMN response at the early stage of processing.The results of Experiment 1 show that the influence of long-term linguistic environment makes/n/-/l/ merged speakers lose their ability to distinguish the merged phonemes in dialects in the early stage of auditory processing.Experiments 2 explored the cognitive processing characteristics and neurological activities of /n/-/l/ in complex contexts between phoneme mergers and non-phoneme mergers,and whether there were differences in the neurological mechanisms of the two groups.The paradigm of Experiment 2 was consistent with that of the experiment 1,and the same group of participants were used.But the sound stimulation in Experiment 2 is not the pronunciation of single words,but the pronunciation of idioms and pseudowords.ERP results show that both merged speakers and unmerged speakers exhibit significant MMN responses,and there is no difference between the amplitudes of MMNs.It seems that the semantic information contained in idioms helps phoneme mergers regain the ability to distinguish/n/and/l/.unmerged speakers exhibited significant late components in 280-340 ms and 440-480 ms,while phonemic mergers did not induce similar components in these two time windows,showing only a later late component in 600-650 ms.The results of Experiment 2 show that high-level cognitive processes,such as semantics,have a top-down effect on the process of /n/-/l/ phoneme processing,helping merged speakers acquire the ability of automatic recognition of / n/-/l/ in early speech.The experiment 3 is a behavioral experiment,which evaluates whether the subjects can distinguish/n/-/l/ phoneme through three behavioral tests with different difficulty.According to the average correct rate of the three groups of behavioral tests,the participants are divided into merged speakers group and unmerged speakers group.Experiment 3 had three tasks: word condition,idiom condition and multiple-stimulus condition.After the behavioral test,the participants' linguistic background was investigated,and their phoneme merger experience was evaluated.Then the participants were divided into two groups: merged speakers and unmerged speakers.Comparing the accuracy rate and reaction time of two groups of participants,it was found that the accuracy rate of unmerged speakers was significantly higher than that of merged speakers in the three tasks,but there was no significant difference in reaction time between the two groups.The results of Experiment 3 show that the ability of merged speakers to distinguish merged phonemes /n/-/l/ is weaker than that of unmerged speakers,which provides behavioral evidence for the phenomenon of phoneme merger.