In the current study, each sentence stimulus was divided into two frames—a sentence frame and a target frame. Sentence frames could build up a context for people to predict the sentence final word, or the target frame. The target frame was a disyllabic noun and served as the critical word in this experiment. Two types of target frames were manipulated: an expected word (the high-cloze condition), and an unexpected but contextually plausible word (the low-cloze condition). We also manipulated the presented language of the sentence frame and of the target frame.
Half of the sentence frame was presented in Mandarin (the Mandarin mode) whereas the other half was in Taiwanese (the Taiwanese mode). The language of the target frame was either kept the same as the sentence frame (the non-switched condition), or switched into another language and formed a code-mixed sentence (the switched condition). Eight experimental conditions were thus created: high-cloze Mandarin non-mixed sentences (H-MM), high-cloze Mandarin-Taiwanese code-mixed
sentences (H-MT), high-cloze Taiwanese-Taiwanese non-mixed sentences (H-TT), high-cloze Taiwanese-Mandarin code-mixed sentences (H-TM), low-cloze Mandarin non-mixed sentences (L-MM), low-cloze Mandarin-Taiwanese code-mixed sentences (L-MT), low-cloze Taiwanese-Taiwanese non-mixed sentences (L-TT), and low-cloze Taiwanese-Mandarin code-mixed sentences (L-TM). Table 2 summarized a set of example stimuli for each condition. A total of 232 sentences were used in the main experiment. The 232 sentences were counterbalanced across the eight experimental conditions. Eight experimental lists were constructed such that no sentence context or target was repeated, either in Mandarin or in Taiwanese, within the same list. The presentation order of the sentence stimuli was randomized within each list.
Participants were randomly assigned to one of the eight lists.
Table 2: A Set of Example Stimulus
Conditions Cloze
H-MM High Mandarin Mandarin Gemi dou xiagyao nadao ni de qianming
H-MT High Mandarin Taiwanese Gemi dou xiagyao nadao ni de tshiam-miâ
H-TT High Taiwanese Taiwanese Kua-bê long siūnn-beh thê-tio̍h li ê tshiam-miâ
H-TM High Taiwanese Mandarin Kua-bê long siūnn-beh thê-tio̍h li ê qianming
L-MM Low Mandarin Mandarin Gemi dou xiagyao nadao ni de zhuzhi
L-MT Low Mandarin Taiwanese Gemi dou xiagyao nadao ni de tsū-tsí
L-TT Low Taiwanese Taiwanese Kua-bê long siūnn-beh thê-tio̍h li ê tsū-tsí
L-TM Low Taiwanese Mandarin Kua-bê long siūnn-beh thê-tio̍h li ê zhuzhi
To manipulate the expected/unexpected effect, the sentence frames should be able to lead to a particular word that was preferred by a majority of people. The percentage of a particular word used to complete a sentence frame was the cloze
speakers, none of whom participated in the main ERP experiment. Mandarin sentence frames (342 in total) were equally divided into four lists, with each list completed by an average of 30 participants (range 27 to 34). Participants were instructed to read each sentence frame, and to write down three completions based on the order that came to their mind.5 While computing the cloze probability of a target word, only participants’ first answer was taken into consideration. Sentence frames that did not lead to a specific target word were excluded (about 100 sentence frames were deleted at this stage). For the remaining sentence frames, the most preferred completion was chosen as the high-cloze target word. The cloze probability of the high-cloze targets was 69% on average (range: 39% to 100%). As for the low-cloze condition, the target words were either generated by the experimenter or selected from the two additional answers provided by the participants in the norming procedure. Six sentences were exceptions because their low-cloze targets were chosen from the participants’ first answer. The average cloze probability for the low-cloze condition was 0.1% (range:
0% to 10%).
To ensure that the frequency of target words would not bias the results, the frequency of low-cloze words and that of high-cloze words were obtained from the Word List with Accumulated Word Frequency in Sinica Corpus (http://elearning.ling.sinica.edu.tw/CWordfreq.html). Statistic analysis revealed no significant differences in frequency between the two types of cloze probability in Mandarin (high-cloze targets: 319; low-cloze targets: 289; t(239)=.804, p=.422).
To ensure that the low-cloze targets were all contextually plausible, another pilot test was conducted. The experimental sentences were divided into two lists.
5 The constraint of some sentences may be too strong to be completed by three different answers. In these cases, participants were instructed to provide two completions, and leave the third column with an “X” mark.
Within each list, half of the sentences were completed by high-cloze words and the other half were completed by low-cloze words. The sentence frames were not repeated within each list. Each list contained 320 sentences, including the 240 experimental sentences and 80 semantically anomalous filler sentences. A separated group of Mandarin native speakers was recruited. Their task was to judge the plausibility of each sentence on a 5-point scale based on their intuition (1: very implausible, 5: very plausible). Each list was rated by 21 participants. Only sentences that scored above three were considered acceptable and were included into the main ERP experiment (high-cloze targets: 4.41; low-cloze targets: 4.01). Six sentences were deleted at this stage.
After the high-cloze word and low-cloze word for each sentence frame were set, the sentences were translated from Mandarin to Taiwanese by the experimenter (native Mandarin-Taiwanese speaker). In translation, sentence structures were remained in parallel between the two languages. An online Taiwanese dictionary (http://twblg.dict.edu.tw/holodict_new/index.html), which was developed by the Ministry of Education in Taiwan, was consulted during translation. The translation was proofread by two native Mandarin-Taiwanese speakers.
Since Mandarin and Taiwanese were tone languages, and the experimental materials would be presented aurally, tone sandhi rules in Mandarin and in Taiwanese6 were taken into consideration while selecting sentence materials. At final
6 In Mandarin, a third tone would be changed into a second tone when it was followed by another third tone. In the current study each sentence stimulus was composed of a sentence frame and a target frame. If the characters in the frame boundary (i.e., the last character of the sentence frame and the first character of the target frame) were both of a third tone, the former should become a second tone. Due to the fact that we would like to keep the sentence frames identical across different conditions, we avoided the combination of frames that could be influenced by
Mandarin tone sandhi. As for Taiwanese, it is the morphological structure, rather than
check, two sentences were excluded for they involved Mandarin tone sandhi in the frame boundary. The finalized stimuli were 232 sentences (see Appendix I).
The finalized materials were produced by a native Mandarin-Taiwanese male speaker, and recorded in a soundproof room. To keep the sentence frames identical across different conditions (e.g., high-cloze, low-cloze, switched, and non-switched conditions), sentence frames and target frames were recorded separately. Sentence frames, each containing 10 to 15 characters/syllables, were produced word by word.
A pause was added between words during recording. Audacity 1.3, a sound-editing software, was used to make the inter-word pause a 200-ms silence. As for the target frames, the duration was set 1200 ms (i.e., 600 ms for the target word and the remaining 600 ms for silence). The long target frame duration allowed us to examine related ERP components at a late time window, for in the current study EEG data was time-locked to the onset of the target frame. To ensure that participants were listening to the sentences attentively, twenty percent of the sentences were followed by a probe word. Half of the probe words were drawn from the sentence contexts (but never the sentence-final target words), and the other half were disyllabic words not appearing in the sentence stimuli. All of the probe words were content words, and the language of the probe word followed the language of its sentence frame. The duration of probe words was 600 ms.