Abstract
BACKGROUND:Researchers have done much work to investigate semantic priming with event-related potentials (ERPs) method. The ERPs component of N400 is of great importance in this research domain. N400 is a negative wave occurs at about 400 ms after the stimulus onset. It has been accepted that N400 represents the processing of semantic information. In many studies, amplitude of N400 could be reduced by semantic priming. Relatively, ERPs studies on phonological priming, especially by auditory stimuli, deserve to be further investigated.

OBJECTIVE: To investigate the EPR characteristics of semantic and phonological priming with spoken two-character Chinese words, and also to testify theories about auditory word recognition.

DESIGN: Repeated measurement experiment.

SETTING: Center for Psychology and Behavior Studies, Tianjin Normal University.

PARTICIPANTS: This experiment was carried out between August and October 2003 in Tianjin Normal University. Seventeen healthy college students (8 male and 9 female, age ranging from 19 to 23 years) with no hearing defect were involved in this experiment. All of them were Chinese native speakers. All except one male participant were right handed. Informed consents of detected items were obtained from all the participants.

METHODS: In the study, lexical dicision task which required participants was used to judge if the latter words were real words or pseudowords in the auditorily presented word-pairs by pressing buttons. The stimuli materials were 640 two-character word-pairs including semantic related, initial phonological overlap, final phonological overlap, phonological and semantic unrelated and word-pseudoword (control) conditions. These five kinds of word-pairs presented randomly in the experiment. The presentation of stimuli and recording of behavioral data were realized by STIM software. The error rate and mean reaction time (RT) of reaction of subjects were calculated under four experimental conditions. The word-pseudoword items were not included in the analysis (just served as fillers). EEG were recorded referring to linked mastoids with Scan 4.3 software made by Neuroscan company in America. Participants wore the Quick-Cap32 electrode cap (International 10/20 System). EEG data were amplified and acquired with a sampling rate of 500 Hz. Offline exclusion of occular artifacts was done on the acquired continuous EEG data after experiment. Then the EEG data were segmented to epochs of 1 700 ms, each epoch was time-locked to the onset of target stimulus with a 200 ms pre-target baseline. Then the operation of baseline (with a 200ms pre-target baseline), artifact rejection (epochs with amplitude exceeded the range of －100 μV to +100 μV were rejected) and average were realized.

MAIN OUTCOME MEASURES: Error rate, reaction time and caused N400 waveform change of target words under different experimental conditions.

RESULTS: Data from 17 subjects were involved in the final analysis. The error rate of subjects under semantic related condition was lower (P < 0.05) while RT was shorter (P < 0.01) than that under phonological and semantic unrelated condition. The error rates of subjects under initial phonological overlap and final phonological over lap conditions were higher than that under phonological and semantic unrelated condition (P < 0.01). The error reactions were discarded and the mean value of RT under 4 conditions (semantic related, initial phonological overlap, final phonological overlap and phonological and semantic unrelated) were 830 ms, 891 ms, 864 ms and 903 ms, respectively. There were significant differences of error rate and RT between semantic related and phonological and semantic unrelated conditions (P < 0.01). Error rate and RT under final phonological overlap condition were significantly different from those under initial phonological overlap and phonological and semantic unrelated conditions (P < 0.01). Since the largest amplitude of N400 occurred at FCZ electrode, FCZ was chosen for analyzing the effects under four different experimental conditions. The LSD multiple comparison showed that amplitude under control condition was significantly higher than that under semantic related condition (P < 0.01). Latency under the control condition was the longest, then decreased gradually according to the sequence of semantic related, final phonological overlap and initial phonological overlap condition. As the results of LSD multiple comparison revealed, initial phonological overlap condition had a tendency of shortened latency, with significant difference when compared with control condition (P < 0.05). Another five electrodes were selected besides FCZ standing for anterior, parietal, posterior, left and right brain area in order to investigate the scalp distribution difference of N400 under different experimental conditions. Two factors repeated measured ANOVA (Analysis of variance) was carried out for amplitude analysis, with different experimental conditions (4 levels) as the first factor and electrode positions (6 levels) as the second factor. The main effect of electrode positions was significant [F(5,80)= 13.376, P =0.000]. Among the chosen electrodes, FCZ had the largest amplitude (－7.946 μV), the second was CZ (－7.285 μV) and the other top-down sequence was FZ (－6.849 μV), PZ (－4.324 μV), T8 (－3.501 μV ) and T7 (－3.484 μV), respectively. Results of LSD multiple comparison revealed that the amplitude of FCZ was higher than all other electrodes but CZ.

CONCLUSION: Two-character Chinese words presented auditorily in lexical decision task. Semantic relationship can facilitate the recognition of target words, which is shown by lower error rate, shorter RT and decreased amplitude of N400 component. Phonological congruence between prime and target words has both facilitory and interferential effects on target words, which is shown by increased error rate, shorter RT and the tendency of decreasing N400 amplitude. Interferential effect of initial phonological overlap between prime and target words is stronger than that of final phonological overlap. This finding provides evidence for initial phonological information's special function during auditory word recognition which is supported by Cohort Theory. No evidence for the existence of separate intracranial sources of semantic priming and phonological priming has been found. However, Cohort Theory should be revised while applying to the recognition of two-character words of Chinese.

INTRODUCTION

Priming effect means the processing of a word stimulus has a facilitory effect on a subsequent word, given that the latter shares some similarities with the former one. Study of semantic priming showed that semantic relationship between two successively presented words (e.g., "car-road" ) can improve the processing of the latter word [1]. It is so-called semantic priming. The phenomenon of semantic priming has been thoroughly investigated in numerous studies, the stimuli were presented sometimes auditorily, sometimes visually or even presented by a crossmodal way [2-4]. Phonological priming has also been reported in some studies, where phonological congruence or similarity between priming and target words could improve the processing efficiency of target words. Consistent results of priming effect have been reported in the experiments which final phonological congruence words were used as stimuli [5], but such results did not exist in experiments that initial phonological congruence was used. In some cases processing of a word has a facilitory effect on a subsequent word, in other cases the effect is opposite[6], or even no effects at all [7-10].
Domestic researchers have done some studies to investigate semantic priming and phonological priming. Mandarin was used as experimental material in these studies. In Peng Dan-ling's study[11], degree of semantic relationship was controlled. Some word-pairs had strong semantic relationship (such as "prison-jail" ), some had moderate relationship (such as "field-green" ) and also some word-pairs had no semantic relationship (such as "arrest-germ" ). Half of the target words were of high frequency while the other half was low. Participants were asked to perform a lexical decision task. In the results they found that both the strength degree of semantic relationship and the frequency of words had significant effects on reaction time (RT), and reciprocal effect between semantic relationship and frequency was also significant. Recognition of low frequency words was facilitated obviously by semantic relationship, but no effects were observed in high frequency words. With primed naming task, Zhang et al [12] investigated the effect of phonological priming and morphological priming on pronunciation of Mandarin. The result showed that only phonological priming effect was significant.
Researchers have done much work to investigate semantic priming with event-related potentials (ERPs) method. The ERPs component of N400 is of great importance in this research domain. N400 is a negative wave occurs at about 400 ms after the stimulus onset. It has been accepted that N400 represents the processing of semantic information. In many studies, amplitude of N400 could be reduced by semantic priming [13-17]. There are relatively few domestic ERPs studies on phonological priming, especially by auditory stimuli [18]. Actually, it will be valuable for auditory language recognition by investigating through the method of presenting stimuli auditorily. Cohort Theory suggested by Marslen reflects that word recognition strictly follows a sequence. The initial imformation of words is most important because inetetmines the word-initial cohort. For instance, when we hear an incomplete sentence of "Paul got a job at the ca-", all possible words might be activated, such as "candy", "cash", "candle", "cashier", "camp". And the quantity of possible words will be reduced and finally one will be focused on with the process of getting more phonological information. However, auditory word recognition is considered by the Trace Model as the matching process between auditory imput and target representation in the mental lexicon. It is an effect of present competitive situation which means that the closer the auditory imput is getting to the representation, the stronger inhibitory efffect it will be on its competitory words. This model mainly concerns the global goodness of fit between auditory imput and mental lexicon while comparing with the Cohort Model which focuses on the importance of initial imformation for auditory language processing[19]. Besides, most studies used single-character words as stimuli in the past, whatever it was for semantic priming or for phonological priming. Large quantity of two-character words is one of the characteristics of Mandarin. So it can be supposed that research on semantic priming of two-character words will be especially useful. This study investigated semantic priming, phonological priming of two-character words and relationship between them using ERPs method. This study mainly concerns two aspects. One is to investigate the characteristics of semantic and phonological priming, respectively. The other is to testify the validity of Cohort Theory and Trace Theory in auditory language recognition. Two experimental conditions with phonological and semantic unrelated and semantic related word-pairs were set in the study. Furthermore, we set another two conditions with initial phonological overlap word-pairs and final phonological overlap word-pairs. If the priming effect under initial phonological overlap condition is different from that under final phonological overlap condition, the Cohort Theory will be approved.

SUBJECTS AND METHODS

Subjects
This experiment was carried out between August and October 2003 in Tianjin Normal University. Seventeen healthy college students (8 male and 9 female, age ranging from 19 to 23 years) with no hearing defect were involved in this experiment. All of them were Chinese native speakers. All except one male participant were right handed. Informed consents of detected items were obtained from all the participants.

Methods
Stimuli
The stimuli consist of 640 word-pairs with each word having two Chinese characters. Five kinds of primes (five different priming conditions with different relationship within the word-pairs) were used: semantic related primes ( "qi'che-dao'lu" ). Initial phonological overlap primes ( "bi'ye-bi'ran" ). ③final phonological overlap primes ("lan'tian-zeng'tian"). ④controlled primes which were both semanticly and phonologically unrelated to the targets ( "tong'xin-sun'shi" ). The number of items (word-pairs) in each group above was 80. 320 pairs of word-pseudoword items (replace one character in the two-character words to make pseudowords) (the first word is a real word while the second one is a pseudoword, e.g. "ke'zhi-ao'wo" ) served as fillers. All the real two-character words were commonly used and were presented only once to avoid repetition effect. No phonological association between semantic related word-pairs was permitted. That means neither the phonation letter congruence nor the spelling letter congruence exists. And as the same rule, no semantic association for the phonological congruence words (initial overlap or final overlap) was allowed. And of course there was no semantic or phonological association in the unrelated items. Besides, there was no obvious difference in frequency and duration of stimuli presentation. The above stimuli were presented at random order during the experiment.
All the words were read by a female voice and recorded in the PC through 16 bit A/D conversion, with the sampling rate of 20 000 Hz. The audio data were edited by software so as to be equally presented in left and right channels. There was a 600 ms silence interval between the offset of the prime and the onset of the target.

Procedure
Participants were tested individually in a sound-proof screening room, with a moderate lighting condition. Firstly, they received a practice to ensure that the requirements of this experiment were really understood. Instruction was told as the following: "You will hear a list of word-pairs with each word making by two characters. The first item of the pair will always be a real word, but the second one could be a word or pseudoword (there is no such word in Mandarin). Your task is to respond by pressing a button on the response pad. If the second item is a real word, press the first button, otherwise, press the forth button. Please respond as quickly and accurately as possible." They were told to maintain a relax, natural posture and move as little as possible during the experiment.
The stimuli were presented through earplugs with the sound of 90 dB. Presentation of stimuli was controlled by STIM software of NeuroScan, with a 4 s SOA between the onset of two consecutive pairs. The whole experiment lasted 50 minutes with two short-time breaks.

Data recording and analyzing
The presentation of stimuli and recording of behavioral data were realized by STIM software. Only the benavioral data of targets were analyzed by computing participants' error rate and mean reaction time under four experimental conditions, including semantic related, initial phonological overlap, final phonological overlap and phonological and semantic unrelated conditions. EEG data were amplified and recorded referring to linked mastoids and with a sampling rate of 500 Hz by Scan 4.3 software. Participants wore the Quick-Cap32 electrode cap (International 10/20 System) made by Neuroscan company in America.
Offline exclusion of occular artifacts was done on the acquired continuous EEG data after experiment. Then the EEG data were segmented to epochs of 1 700 ms, each epoch was time-locked to the onset of target stimulus with a 200 ms pre-target baseline. Then the operation of the baseline (with a 200 ms pre-target baseline), artifact rejection and average were realized. The ERP data were analyzed by computing the mean amplitude in selected latency windows, within each condition for each subject. All electrodes except EOG were used as the criterion of artifact rejection, epochs with amplitude exceeded the range of －100 μV to +100 μV were rejected. The rejection rate was less than 15%. Word-pairs containing pseudowords were not analyzed.

Statistical analysis
The statistical analysis were operated by the first author with the software of SPSS 9.0 which contained Chi-square test of error rate and ANOVA (Analysis of variance) of respone time. The EEG date were statistically analyzed by one-way ANOVA and two factors repeated measured ANOVA. Greenhouse-Geisser correction was applied.

RESULTS

Quantitative analysis of the participants
Data from 17 participants were involved in the final analysis.
Behavioral data
The error rate of subjects under semantic related condition was lower (P < 0.05) while RT was shorter (P < 0.01) than that under phonological and semantic unrelated condition. The error rates of subjects under initial phonological overlap and final phonological over lap conditions were higher than that under phonological and semantic unrelated condition (P < 0.01) (Table 1). The error reactions were discarded and the mean value of RT under 4 conditions (semantic related, initial phonological overlap, final phonological overlap and phonological and semantic unrelated) were 830 ms, 891 ms, 864 ms and 903 ms, respectively. There were significant differences of error rate and RT between semantic related and phonological and semantic unrelated conditions (P < 0.01). Error rate and RT under final phonological overlap condition were significantly different from those under initial phonological overlap and phonological and semantic unrelated conditions (P < 0.01).

ERP data
EEG data corresponding to RT which were longer than 1 500 ms or shorter than 200 ms and incorrect responses were discarded from the analysis. ERPs were averaged within each condition so that grand average of 17 participants was obtained (Figure 1).

ERPs under different experimental conditions
A negative component peaking at 400 ms (namely N400) was observed under 4 experimental conditions, each with different amplitude and latency. Since the largest amplitude of N400 occurred at FCZ electrode (Figure 1d ), FCZ was chosen for analyzing the effects under four different experimental conditions(Figure 2). Amplitude and peak-latency (from the onset of stimulus to the peak within the time window of 300-500 ms) of N400 were measured at FCZ, within each condition for each participant. The data were shown in Table 2. Analysis of variance (ANOVA) was carried out for amplitude data. The LSD multiple comparison showed that amplitude under control condition was significantly higher than semantic related condition (P < 0.01). Latency under the control condition was the longest, then decreased gradually according to the sequence of semantic related, final phonological overlap and initial phonological overlap condition. As the results of LSD multiple comparison revealed, initial phonological overlap condition had a tendency of shortened latency, with significant difference when compared with control condition (P < 0.05).

Comparison of the difference of scalp distribution of N400 potential under different conditions
Another five electrodes were selected besides FCZ standing for anterior, parietal, posterior, left and right brain area in order to investigate the scalp distribution difference of N400 under different experimental conditions. Two factors repeated measured ANOVA was carried out for amplitude analysis, with different experimental conditions (4 levels) as the first factor and electrode positions (6 levels) as the second factor.Main effects of four experimental conditions were not significant [F(3,48)=2.769, P =0.067], obviously this was caused by the additional electrodes except FCZ. The main effect of electrode positions was significant (F(5,80)= 13.376, P =0.000), and there was no significant reciprocal effect [F(15,240)=1.819, P =0.122].

Among the chosen electrodes, FCZ had the largest amplitude (－7.946 μV), the second was CZ (－7.285 μV) and the other top-down sequence was FZ (－6.849 μV), PZ (－4.324 μV), T8 (－3.501 μV) and T7 (－3.484 μV), respectively. Results of LSD multiple comparison revealed that the amplitude of FCZ was higher than all other electrodes but CZ. The amplitude of CZ was significantly higher than that of PZ, T7 and T8, but no significant difference was found between CZ and FZ. Differences between PZ and T7, PZ and T8 were not significant either.

DISCUSSION

Since semantic relationship between priming and target words caused lower error rate and shorter RT in lexical decision task, the results of this study showed clear evidence of semantic priming. It has been testified that priming effect can occur under the condition that two-character Chinese words were auditorily presented. Revealed by ERPs data semantic related condition were associated with smaller N400 than control condition. The results were consistent with those obtained by some overseas researchers.
In some studie whichs priming effect was investigated incorrect responses was discarded so as to analyze those correct responses. In our opinion, it's necessary to include incorrect responses in the analysis. In this study, error rates under initial and final phonological overlap conditions were obviously higher than control. It could not be regard as just a chance.
If there are only two conditions in a priming experiment and the priming conditions cause higher error rate and shorter RT than the control condition, maybe it will be difficult to decide whether the shorter RT is caused by priming effect or by different standards of responses adopted by participants. In this case, decreased error rate and RT were found under semantic related condition, so the possibility of different standards of responses were adopted could be excluded.
The tendency of shorter RT was observed in final phonological overlap condition compared with control condition. It may suggest that final phonological overlap has a facilitory effect on word recognition. The amplitude of N400, which was consistent with behavioral data, showed a tendency of decreasing under final phonological overlap condition at FCZ, compared with control condition. But the higher error rate under final phonological overlap condition may suggest that final phonological overlap also has an interferential effect. So, final phonological overlap may have double effects. Priming words have both facilitory and interferential effects on target words.
In this study, error rate was higher under initial phonological overlap condition compared to control condition, but the shortening of reaction time was not obvious. With these results we can conclude that the overlap of initial phonology between prime and target words mainly has interferential effect on words recognition. The condition of final phonological overlap was quite different. The interferential effect of initial phonological were stronger in both conditions using two-character Chinese words. Also we consider that initial overlap of phonology between prime and target word probably has a facilitatory effect, but the interferential effect is so large that facilitatory effect is covered. From the ERPs data under initial overlap condition we can see, amplitude of N400 shows an inclination of declining at FCZ, thus possibly providing further evidence for above conclusion. According to the behavioral results of this study, stronger interferential effect happened under the condition of initial phonological overlap than the condition of final phonological overlap. Forthermore, ERP data showed that N400 peak-latency got a tendency of occurring ealier which seemes as an interesting result. To some certain this result approve the hypothesis that initial phonological information plays a special role on auditory word recognition which is part of Cohort Theory. However, since the initial phonological overlap between primes and targets showed a stronger interferential effect in our study, the result is not completely consistent with Cohort Theory. Results of our study implys that when processing two-character words with initial phonological overlap the first character might be processed sooner. But the stronger misguidance effect given by the first character during processing results in the longer RT comparing with the final phonological overlap condition. So it illuminates that Cohort Theory should be revised while applying to the recognition of two-character words of Chinese.
According to the behavioral results of this study, semantic priming caused the greatest shortening of RT, the degree of such shortening was slightly small in phonological priming. As for ERPs data at FCZ, semantic priming corresponded to much lower N400 amplitude, meanwhile effect was found in phonological priming. Such findings are consistent with overseas conclusions [18]. No difference in scalp distribution was found for semantic and phonological priming, therefore no evidence for separate intracranial sources could be provided. Considering the fact that relative few channels were used for EEG data recording in this study, It is necessary to use more channels (such as 128 or 256 channels), specific analysis technology (such as using software of Source or Curry) or some imaging technologies (such as fMRI or PET) to obtain reliable conclusions.However, our result is consistent with other studies by finding that N400 elicited by auditory stimuli is close to centro-frontal area which is different from that elicited by visual stimuli which peak at parietal[20,21].
As Figures 1 and 2 revealed, no significant components of N1 or P2 were observed. This is different from some other studies [13,18]. Connolly et al[20,21] have ever found that auditory stimuli induced smaller N1, P2 than visual stimuli. It is probably because there is no obvious limit between auditory words in spoken language. In this study a 600 ms interval between prime and target words was used as the unambiguous limit. However, the same absence of basic waveforms in such condition is still a problem need to be solved. Of course, the stimuli and the presentation methods of stimuli are different from other studies, maybe these factors can cause the absence of these early sensory components. The meaning and phonology of two-character words are more complicated than those of single-character words. Maybe basic waveform will occur by increasing the interval between prime and target words. Further work should be done to test this assumption.
Two-character Chinese words were presented auditorily in lexical decision task. Semantic relationship can facilitate the recognition of target words, which is shown by lower error rate, shorter RT and decreased amplitude of N400 component. Phonological congruence between prime and target words has both facilitory and interferential effects on target words, which is shown by increased error rate, shorter RT and the tendency of decreasing N400 amplitude. Interferential effect of initial phonological overlap between prime and target words is stronger than that of final phonological overlap. This finding provides evidence for initial phonological information's special function during auditory word recognition which is supported by Cohort Theory. No evidence for the existence of separate intracranial sources of semantic priming and phonological priming has been found. However, Cohort Theory should be revised while applying to the recognition of two-character words of Chinese.
Acknowledgement: The study here acquired the help and guidance from Wei Jinghan, who is a researcher from Institute of Psychology, Chinese Academy of Sciences. We are deeply grateful to Professor Wei in this article.

REFERENCES

Tivarus M E, Ibinson G W, Hillier A, et al. An fMRI study of semantic priming: Modulation of brain activity by varying semantic distances. Cogn Behav Neurol 2006;19(4):194-201
Swinney D A, Onifer W, Prather P, et al. Semantic facilitation across sensory modalities in the processing of individual words and sentences. Mem Cognit 1979;7(3):159-165
Friedrich M, Friederici AD. Phonotactic knowledge and lexical-semantic processing in one-year-olds: brain responses to words and nonsense words in picture contexts. J Cogn Neurosci 2005;17(11):1785-1802
Kreher DA, Holcomb PJ, Kuperberg GR. An electrophysiological investigation of indirect semantic priming. Psychophysiology 2006;43(6): 550-63
Emmorey KD. The neurological substrates for prosodic aspects of speech. Brain Lang 1987;30(2):305-320
Radeau M, Morais J, Dewier A. Phonological priming in spoken word recognition: task effects. Memory Cognition 1989;17(5):525-535
Slowiaczek LM, Hamburger M. Prelexical facilitation and lexical interference in auditory word recognition. J Exp Psychol Learn Mem Cogn 1992;18(6):1239-1250
Slowiaczek LM, Soltano EG, Wieting S J,et al. An investigation of phonology and orthography in spoken-word recognition. Q J Exp Psychol A 2003;56(2):233-262
Bonte M, Blomert L. Developmental changes in ERP correlates of spoken word recognition during early school years: a phonological priming study. Clin Neurophysiol 2004;115(2):409-423
Slowiaczek LM, Soltano EG, Bernstein HL. Lexical and Metrical Stress in Word Recognition: Lexical or Pre-lexical Influences? J Psycholinguist Res 2006;35(6):491-512
Zhu Y. Experimental Psychology. Beijing: Beijing University Press. 2000:410-412
Zhang HC, Shu H, Phonological priming and orthography priming in Chinese words. Xinli Xuebao 1989;3:284-289
Luo YJ, Hu S, Weng XC, et al. Effects of semantic discrimination of Chinese words on N400 component of event-related potentials. Percept Mot Skills 1999;89(1):185-193
Deacon D, Dynowska A, Ritter W. Repetition and semantic priming of nonwords: implications for theories of N400 and word recognition. Psychophysiology 2004;41(1):60-74
Hill H, Ott F, Weisbrod M. SOA-dependent N400 and P300 semantic priming effects using pseudoword primes and a delayed lexical decision. Int J Psychophysiol 2005;56(3):209-221
Matsumoto A, Lidaka T, Haneda K. Linking semantic priming effect in functional MRI and event-related potentials. Neuroimage 2005;24(3): 624-634
Matsumoto A, Lidaka T, Nomura M. Dissociation of conscious and unconscious repetition priming effect on event-related potentials. Neuropsychologia 2005;43(8):1168-1176
Radeau M, Besson M, Fonteneau E, et al. Semantic, repetition and rime priming between spoken words: behavioral and electrophysiological evidence. Biol Psychol 1998;48(2):183-204
Song FN, Ma RJ. Two Models of Auditory Word Recognition.Psychol Sci 2001;24(6):746
Connolly JF, Stewart SH, Phillip NA. The effects of processing requirements on neurophysiological responses to spoken sentences. Brain Lang 1990;39:302-318
Connolly JF, Phillips NA, Stewart SH, et al. Event-related potential sensitivity to acoustic and semantic properties of terminal words in sentences. Brain Lang 1992;43:1-18

听觉呈现条件下汉语双字词语义和语音启动的事件相关电位研究**☆

吕 勇，杜英春，宋 娟，沈德立
天津师范大学心理与行为研究中心，天津市 300074
吕 勇☆，男， 1963年生，天津市人，汉族，博士，教授，主要从事认知神经科学（事件相关电位）研究。
通讯作者：沈德立，教授，博士生导师，天津师范大学心理与行为研究中心 天津市 300074
教育部人文社会科学重点研究基地重大项目(05JJDXLX003)*；天津市科技发展计划项目（05YFGDGX10200）*
摘要
背景：在关于语义启动的事件相关电位研究中，N400（在刺激呈现后400 ms左右出现的负波）成分占据重要的地位，被普遍认为反映了语义加工过程。不少研究证实，语义启动会导致N400波幅降低。而语音启动的研究较少，尤其是通过听觉呈现刺激的方式进行的的事件相关电位研究。
目的：观察听觉呈现条件下汉语双字词的语音启动和语义启动事件相关电位特征，验证关于听觉词汇识别的相关理论。
设计：重复测量实验。
单位：天津师范大学心理与行为研究中心。
对象：实验于2003－08/10在天津师范大学进行。选择17名天津师范大学在读大学生为被试，男生8名，女生9名，年龄19～23岁，身体健康，听力正常，母语为汉语，普通话标准。除1名男生外，其余均为右利手。所有受试大学生均对检测项目知情同意。
方法：①本实验采用词汇决定任务，判断继时听觉呈现的词对中，后一个词为真词还是假词。要求被试进行按键反应。刺激材料为640对双字词组成的词对（包括词义相关、首音相同、尾音相同、音义无关及真假词对5类）。实验中5类词对以随机顺序呈现。②使用Stim呈现实验材料，该软件同时记录行为数据，对被试在4种实验条件下（词对间语义相关、首音相同、尾音相同、音义无关）的反应的错误率和平均反应时进行计算。对真假词对不进行分析（作为填充）。被试头戴美国Neuroscan公司Quick-Cap32导脑电帽（国际10-20系统），参考电极为双侧乳突连线，经Synamps放大器放大15 000倍，采样频率500 Hz，然后送入电脑由Scan 4.3软件记录。对数据的离线分析包括利用SCAN软件中的Ocular Artifact Reduction功能去除连续文件中的眼电干扰、分段（分析时程从触发点前200 ms至触发点之后1 500 ms）、基线矫正 （－100～0 ms为标准）、去伪迹（超过±100 μV的数据作为伪迹排除）、叠加。
主要观察指标：不同实验条件下目标词进行反应时的错误率、反应时间及引起的N400波形的变化。
结果：17名被试的数据均被纳入结果分析。①语义相关条件下被试错误率低于音义无关（P < 0.05）,反应时短于音义无关（P < 0.01）；首音相同及尾音相同条件下被试错误率高于音义无关（P < 0.01）。将错误的反应剔除，语义相关、首音相同、尾音相同和音义无关4种条件下的平均反应时分别为830，891，864和903 ms。语义相关与音义无关相比差异显著（P < 0.01）；尾音相同与音义无关及首音相同差异明显（P < 0.01）。②因为FCZ引发的N400最为显著，所以对4种实验条件下FCZ点引发的N400的波幅和潜伏期进行统计分析。LSD多重比较发现，音义无关波幅显著高于语义相关（P < 0.01）,音义无关峰潜时最长，其次分别为语义相关、尾音相同，时间最短的是首音相同。LSD多重比较显示，首音相同具有峰潜时缩短的趋势，与音义无关差异显著（P < 0.05）。③考察不同实验条件下N400头皮分布的差异，除FCZ点之外，又选取了FZ、CZ、PZ、T7、T8分别作为前区、顶区、后区、左区和右区的代表点，对波幅进行4（实验条件）×2（电极位置）两因素重复测量方差分析。不同电极位置的主效应非常显著（P =0.000）。波幅最高的是FCZ（－7.946 μV），其次分别为CZ（－7.285 μV）、FZ（－6.849 μV）、PZ（－4.324 μV）、T8（－3.501 μV）、T7（－3.484 μV）。LSD多重比较表明，除CZ外，FCZ波幅高于其他代表点。
结论：①在词汇决定实验中，通过听觉通道呈现汉语双字词，启动词与目标词之间存在语义联系可导致目标词的加工得到改善，表现为错误率降低、反应时缩短，以及N400波幅减小。②启动词与目标词之间的语音一致对目标词的加工既有干扰作用，又有促进作用。表现为错误率提高，但反应时趋向缩短，N400波幅亦呈下降趋势。③启动词与目标词首音相同比尾音相同有更大的干扰作用。这为交股理论所认为的在听觉词汇识别过程中首音具有特殊意义的看法提供了支持，不过该理论应用于汉语双字词识别需要进行一定的修正。④本实验没有发现语义启动语音启动具有不同脑内源的证据。
关键词：汉语；启动效应；语义启动；语音启动；事件相关电位
中图分类号:R496 文献标识码:A 文章编号:1673-8225(2007)22-04451-06
吕勇，杜英春，宋娟，沈德立.听觉呈现条件下汉语双字词语义和语音启动的事件相关电位研究[J].中国组织工程研究与临床康复，2007，11(22):4451-4456
[www.zglckf.com/zglckf/ejournal/upfiles/07-22/22k-4451(ps).pdf]
(Edited by Zhao L/Song LP/Wang L)