Facilitation of processing darenimo ‘any/everyone’ negative Japanese sentences using prosodic entrainment

3.1.1 Participants

Thirty-five participants whose native language is Japanese (14 females and 21 males) were recruited as undergraduate students from a university in Aichi prefecture, Japan to perform the acceptability judgment task. Seventeen participants came from prefectures in the Tokai region (13 from Aichi, 3 from Mie, and 1 from Gifu) while others came from various other prefectures in Japan (3 from Hiroshima, 2 each from Osaka and Okayama, 1 each from Fukuoka, Hyogo, Kanagawa, Nagano, Shizuoka, Tochigi, Toyama, and Wakayama). At the time of the questionnaire survey, they were all living in or near the city of Nagoya, Japan. Their ages ranged from 18 years and 5 months to 22 years and 7 months. The average age was 19 years and 1 month with a standard deviation of 1 year. These students voluntarily participated in the questionnaire survey. All collected information was stored in a secure location, and the participants were given numerical pseudonyms to ensure privacy.

3.1.2 Stimulus sentences

Forty pairs (80 sentences in total) of negative and positive polarity sentences with darenimo were created for the acceptability judgment task. Sentence (5a) is an example of negative polarity while sentence (5b) is an example of positive polarity. All sentences were presented in the non-past verb form (see all stimuli in Appendix).

3.1.3 The questionnaire for acceptability judgments

Forty negative polarity and 40 positive polarity sentences were visually presented to each participant in a written questionnaire form. Participants were asked to judge how acceptable each of 80 sentences was as a Japanese sentence by rating them from 1 point (not acceptable at all) to 7 points (very acceptable). To avoid the close placement of paired sentences, all sentences were pseudo-randomized so that paired sentences did not appear in close proximity to each other.

3.2.1 Data for acceptability judgments

A total of 2,795 acceptability judgments (5 were missing values) by 35 participants using the 1–7 scale for the 80 stimuli sentences were analyzed using a linear mixed effect (LME) model (Baayen et al. 2008) and the lme4 package (Bates et al. 2014) within R (R Core Team 2014). The fixed variable was polarity (negative or positive). The random variables were the participants and the sentences (random intercept and slopes). Results from multiple models were compared according to the Akaike Information Criterion (AIC). The final model used was the lmer (accptscore ∼ (1 |participant) + (0 + polarity |participant) + (1|item) + polarity, data).

3.2.2 Results of LME model analyses

Acceptability judgment scores were analyzed with the lmer function using the restricted maximum likelihood (Harville 1977). Satterthwaite’s approximations were applied via the lmerTest package to generate p values for each model (Kuznetsova et al. 2014). The best-fit LME model was chosen based on model comparisons using AIC (Akaike’s Information Criterion compared by the maximum likelihood, Anderson et al. 2000). The means (M), standard deviations (SD), and standard errors (SE) for negative or positive polarity sentences are reported in Table 2. The result of the LME analysis is presented in Table 3.

As shown in Table 3, scores for acceptability judgments on the 1–7 point scale showed a significant polarity effect (t[51.76] = −7.56, p < 0.001, reference of negative polarity). This result indicated that negative polarity expressions with darenimo (M = 6.16, SD = 1.56, SE = 0.04) were perceived to be more acceptable as a Japanese sentence than those with positive polarity expressions (M = 4.23, SD = 1.97, SE = 0.05).

3.2.3 Discussion

Study 2 investigated subjective acceptability judgments for written sentences containing darenimo with polarity. It is assumed that, as a higher frequency of combinations in the corpus are heard and used, these patterns are firmly stored in the mental lexicon of native Japanese speakers. The result of the questionnaire survey of acceptability judgments suggested that darenimo is perceived as highly acceptable in Japanese when it is used in expressions with negative polarity. The result of Study 2 was congruent with the result of corpus frequencies in Study 1, indicating that the less frequently occurring positive polarity expressions with darenimo were judged to be less acceptable than the more frequently occurring negative polarity expressions. Consistent with findings in Study 1, subjective acceptability judgments in Study 2 supported one-sided distribution (Kato 1985) rather than complementary distribution (Gunji 2006). It should also be noted that although the frequency effect is usually discussed at the word level (e.g., Balota and Spieler 1999; Brown and Rubenstein 1961; Gordon 1983; Hino and Lupker 1998; Jescheniak and Levelt 1994; Taft 1979), the present study showed that frequency of darenimo with polarity expressions at the sentence level also affects the degree of subjective acceptability judgments.

4.1.1 Participants

Forty-two undergraduate students whose native language is Japanese (15 females and 27 males) from a university in Aichi prefecture, Japan, participated in the experiment. Thirty-one participants came from prefectures in the Tokai region (22 from Aichi, 6 from Mie, and 3 from Gifu) while the others came from various other prefectures in Japan (2 from Hiroshima, and 1 each from Ehime, Fukui, Kanagawa, Osaka, Okayama, Shizuoka, Tochigi, Toyama, and Wakayama). At the time the experiment was conducted, they were all living in or near the city of Nagoya. Their ages ranged from 18 years and 3 months to 24 years and 3 months. The average age was 19 years and 10 months with a standard deviation of 1 year and 4 months. All participants received monetary compensation in exchange for their participation and provided written informed consent. All collected information was stored in a secure location, and the participants were given numerical pseudonyms to ensure privacy. Participants who took part in Study 2 were not included in Study 3.

4.1.2 Stimulus sentences

The same forty pairs of negative and positive polarity sentences used in Study 2 were used for the prosody-and-polarity matching decision task in Study 3. A female native Japanese speaker speaking standard Japanese read the negative and positive polarity sentences aloud with the appropriate prosody. All 80 sentences were recorded by Praat version 6.0.43 (Boersma and Weenink 1992–2002, retrieved June 8 2019). A correctly matched prosody-and-polarity pair with sound spectrums and pitches for sentences (5a) and (5b) is shown in Figure 2. The upward prosody for sentence (5a), which registered 2,832 Hz at the highest F₀ peak, matched with negative polarity. The downward prosody for sentence (5b), which registered 978 Hz at the lowest F₀ peak, matched with positive polarity (see all stimuli in Appendix).

Figure 2:

A stimulus pair of darenimo prosody and polarity entrainment.

These are correctly entrained items responded to by participants as YES.

Study 3 assumed that darenimo upward prosody associates with negative polarity while downward prosody associates with positive polarity. These matches were combinations to which participants were expected to respond with YES (matched). Forty (5a)-type negative polarity sentences with darenimo upward prosody and 40 (5b)-type positive polarity sentences with darenimo downward prosody were used as the stimulus sentences.

To create incorrect response pair combinations, darenimo sounded with upward prosody for negative polarity in (5a)-type sentences was replaced with darenimo downward prosody for positive polarity in (Appendix 1c)-type sentences. These made up the mismatch combinations to which participants were expected to respond with NO (incongruent). Likewise, darenimo sounded with downward prosody for positive polarity in (5b)-type sentences was replaced with darenimo upward prosody for negative polarity (Appendix 1d)-type sentences. These also comprised the prosody-and-polarity mismatch combinations responded to with NO. Forty negative polarity sentences with darenimo downward prosody, and 40 positive polarity sentences with darenimo upward prosody were used in the experiment.

A 200 ms silent pause was inserted both before and after the sound of darenimo. A total of 160 sentences were prepared. Eighty sentences of (5a)-type and (5b)-type met the prosody-and-polarity matched condition (correct, YES responses) while 80 sentences of (Appendix 1c)-type and (Appendix 1d)-type met the prosody-and-polarity mismatched condition (incorrect, NO responses). In order to ensure that participants would see only one prosody/polarity combination in each sentence per experimental session, the experimental stimuli of 160 sentences were divided into four lists to be assigned to the four different participant groups.

Sound durations were controlled equally across the four conditions of prosody-and-polarity matched YES and mismatched NO responses. The means were 2,394 ms (SD = 233 ms, SE = 37 ms) for (5a)-type sentences, 2,405 ms (SD = 229 ms, SE = 36 ms) for (5b)-type sentences, 2,309 ms (SD = 250 ms, SE = 40 ms) for (Appendix 1c)-type prosody-and-polarity mismatched sentences, and 2,298 ms (SD = 253 ms, SE = 40 ms) for (Appendix 1d)-type prosody-and-polarity mismatched sentences. These means were analyzed by one-way analysis of variance (ANOVA). The result showed no significant differences (F[3, 156] = 2.12, p = 0.099, ns). Thus, sound durations were assumed not to influence the prosody-and-polarity matching decision task. To ensure that no influence occurred, sound durations were also included as an LME fixed factor for the subsequent analysis.

4.1.3 Procedure

The present experiment employed an online measurement task for prosody-and-polarity matching (entrainment) decision processing. A ‘+’ symbol used as an eye fixation was initially presented at the center of the computer screen for 250 ms, followed by a sentence including darenimo with upward or downward prosody phonetically presented through the headphone. The participant was required to decide whether the initially presented darenimo prosody matched the negative or positive polarity of the rest of the sentence. In this task, if the participants thought that the upward or downward darenimo prosody and a negative or positive polarity matched, they were instructed to press the YES key after seeing the question mark ‘?’, but if not (mismatch), they were asked to press the NO key. The ‘?’ appeared just after the sound offset of darenimo to prompt a participant to start responding to a task. The subsequent trial started 300 ms after either the YES or NO key was pressed. The participants were asked to perform the prosody-and-polarity matching task as accurately and quickly as possible. All stimulus sentences were randomly presented for each participant. Before the experiment proper, four practice sentences were given.

4.2.1 Data and analyses of the prosody-and-polarity matching decision task

The reaction times and binomial accuracies data collected by the prosody-and-polarity matching decision task were analyzed using the linear mixed effect models (Baayen et al. 2008) and the lme4 package (Bates et al. 2014) within R (R Core Team 2014). The fixed effects were upward or downward prosody, negative or positive polarity, and trial order. The random effects were participants and stimulus sentences (random intercept and slopes). The data for reaction times consisted only of data from trials with correct judgments. Satterthwaite’s approximations were used via the lmerTest package to generate p values for each model (Kuznetsova et al. 2014) using the restricted maximum likelihoods (Harville 1977). For binomial accuracy data, the glmer function (logit link function) was used to calculate the z distribution by maximum likelihood (Laplace approximation).

4.2.2 Results of LME model analyses for accuracy data

A total of 1,680 responses (42 participants × 40 items) were analyzed. According to model comparisons using AIC (Anderson et al. 2000), the fixed factor of sound duration was removed in the best-fit LME model for accuracy because it was not significant. The final best-fit LME model was glmer (acc ∼ (1|subject) + (1|item) + prosody*polarity + trial.z, data, family = binominal). The means (M), standard deviations (SD), and standard errors (SE) are reported in Table 4.

The result of the best-fit LME model is reported in Table 5. The results indicated that all three fixed factors were significant: prosody (z = 5.56, p < 0.001), polarity (z = −8.62, p < 0.001), and trial (z = 2.30, p < 0.05). Furthermore, the interaction between prosody and polarity was significant (z = −4.75, p < 0.001). To examine the results in detail, accuracies of the four prosody/polarity combinations were compared using the R package of lsmeans (least-squares means, Searle et al. 1980). The means with standard errors are depicted and the comparison results are shown in Figure 3, indicating that upward prosody with negative polarity (M = 90.24%) was the most accurately identified, followed by downward prosody with negative polarity (M = 75.95%). Both upward and downward prosody with negative polarities were perceived more accurately than upward (M = 46.43%) and downward (M = 47.62%) prosody with positive polarity, both of which showed approximately equal accuracies (i.e., not significantly different). The most important result from Experiment 3 was obtained from the prosody-and-polarity matched combinations (YES responses). The combination of darenimo upward prosody with negative polarity was judged most accurately while the combination of darenimo downward prosody with positive polarity was judged accurately only randomly.

Figure 3:

Bar graph for accuracies of prosody-and-polarity (mis)matching.

Values in bars are means and values following ± are standard errors. Bars in black are correctly matched prosody-and-polarity combinations (YES responses) while bars in white are incorrectly matched prosody-and-polarity combinations (NO responses). < and = refer to the results of comparison by least-squares means.

It should also be noted that, as shown in Figure 3, downward prosody with negative polarity was more accurately processed than both combinations of upward and downward prosody with negative polarity. This may have been caused by the fact that negative polarity is usually presented with a phonologically prominent feature (i.e., upward prosody or rising pitch). Because native Japanese speakers have sensitivity to upward prosody entrained with negative polarity, they may have been better able to correctly reject downward prosody associated with negative polarity than upward and downward prosody with positive polarity. This accuracy trend also supports the idea that darenimo with upward prosody strongly entrains with negative polarity.

4.2.3 Results of LME model analyses for reaction time data

Twenty-three items of extremely quick responses (less than 30 ms) and 5 items of extremely slow responses (greater than 10,000 ms) were removed from the original data set of 1,680 responses (42 participants × 40 items). After excluding the outliers, the remaining 1,070 items with correct responses were analyzed for reaction times. Based on the Box–Cox power transformation technique (Box and Cox 1964; Venables and Ripley 2002), a logarithmic transformation (natural log) was applied to the reaction times to attenuate skewness in their distribution. Reaction times were analyzed with the lmer function using the restricted maximum likelihood (Harville 1977). Satterthwaite’s approximations were used via the lmerTest package to generate p values for each model (Kuznetsova et al. 2014).

The fixed factors of prosody and polarity were coded by a contrast of 0.5 and −0.5. The variables were the four fixed factors of prosody, polarity, sound duration, and trial, and two random factors of participant and item. The fixed factor of trial was centralized into z values, coded as trial.z. Sound duration was initially included for analysis, but it was found not to be significant, and did not demonstrate a good model fit, so it was removed from the LME model. According to model comparisons using AIC (Anderson et al. 2000), the best-fit LME model was lmer (log(rt) ∼ prosody + polarity + trial.z + (1|participant) + (1|item), data). After this, potentially influential outliers were removed with standardized residuals exceeding 2.5 standard deviation units of the final best-fit LME model (31 of 1,070 responses were removed). The result of the LME model analysis for 1,039 responses is reported in Table 6. The means (M), standard deviation (SD) and standard errors (SE) for prosody and polarity were calculated and reported in Table 7 after the removal of potentially influential outliers.

As shown in Table 6 and depicted in Figure 4, reaction times for prosody-and-polarity matching decisions were significant for prosody (t[976.16] = −6.75, p < 0.001), but not for polarity (t[990.37] = −0.56, p = 0.58, ns). The factor of trial was also significant (t[995.20] = −1.99, p < 0.05). Moreover, the interaction of polarity and prosody was significant (t[987.96] = 5.55, p < 0.001). To examine the details, reaction times of the four conditions were compared using the R package of lsmeans (least-squares means, Searle et al. 1980). The results of comparisons indicated that upward prosody with negative polarity (M = 573 ms) was recognized faster than the other three combinations of downward prosody with negative polarity (M = 765 ms), and upward (M = 809 ms) and downward (M = 864 ms) prosody with positive polarity. These final three combinations were processed at similar speeds (i.e., not significantly different).

Figure 4:

Bar graph for reaction times of prosody-and-polarity (mis)matching.

Values in bars are means and values following ± are standard errors. Bars in black are correctly matched prosody-and-polarity combinations (YES responses) while bars in white are incorrectly matched prosody-and-polarity combinations (NO responses). < and = refers to the results of comparison by least-squares means.

4.2.4 Discussion

Using stimulus sentences from Study 2, Study 3 examined whether darenimo upward or downward prosody entrains with negative or positive polarity to clarify their entrainment distributions. If processing speed and accuracy showed complementary distribution of darenimo for prosody and polarity, the prosodic feature applied to the polarity feature might be understood to be stored as separate aspects of prosody and polarity (Gunji 2006). However, if the distribution were unbalanced, it would suggest that the way in which a particular prosody attached to the lexical item of darenimo entrains with a certain polarity is stored as an associated lexical memory. The result showed that upward prosody of darenimo for negative polarity, which was the most frequent pattern found in Study 1 and the most highly perceived as natural Japanese in Study 2, was the most accurately and quickly perceived. In contrast, downward prosody of darenimo for positive polarity resulted in random accuracies (slightly lower than 50%) and slower reaction times. Therefore, Study 3 strongly supported the unbalanced distribution of darenimo for prosody and polarity (Kato 1985); that is, the prosodic entrainment strategy functions only when upward prosody is attached to darenimo with subsequent negative polarity.