The first two experiments tested the interpretation of polar questions in British-English, even though the conclusions that we would like to draw are much wider in scope. To address this, we replicated Exp. 2 in a different language: Belgian-French. Of course, even if we confirm the findings of Exps. 1 and 2 in one different language, this would not guarantee that the pattern of results is universal. It would, however, increase the experimental base on which to build more wide-ranging conclusions.
There are a couple of differences between polar questions in English and French that warrant discussion up front. First, French has three constructions that can be used to ask questions. To illustrate, the question ‘Do you dance? can be translated into French as:
Declarative: Vous dansez?
Est-ce que: Est-ce que vous dansez?
In order to facilitate the comparison with English, and because it is the most natural form with which to ask negative questions, we tested questions that are marked by means of subject-verb inversion, as in (5b).
Second, French has no distinction between high-neg and low-neg questions (e.g., ‘Didn’t he call?’ and ‘Did he not call?’). Hence, we only tested two conditions: positive polar questions (‘Did he call?’) and negative polar questions (‘Didn’t he call?’).
Third, the response particle system in French is less ambiguous than in English. In particular, when responding to a negative question, ‘Yes’ and ‘No’ in English are variously used to confirm or reject the corresponding question. In our experiment, however, we always used ‘Yes’ to confirm and ‘No’ to reject questions, and the intended meaning was disambiguated in the continuation (e.g., ‘Yes, he has’ vs. ‘No, he hasn’t’). In French, by contrast, ‘Si’ unambiguously confirms and ‘Non’ unambiguously rejects negative questions (Noveck et al., 2021). Since we are only interested in the question phase, however, this difference is not particularly relevant to our purposes.
In the next section, we describe the experiment in more detail.
42 native French speakers were recruited from Université Libre de Bruxelles via an online pool of participants. They all had uncorrected or corrected to normal vision. They were paid four euros for their participation.
The 42 question–answer pairs that were tested in Exps. 1 and 2 were translated into French and recorded by a native speaker of Belgian-French. Since French does not have the genitive before its argument (e.g., ‘his father’s shirt’ vs.‘la chemise de son père’), we instead placed a temporal adverb between the verb and the object. Participants thus listened to sequences such as:
‘Jean a-t-il repassé cet après-midi sa chemise?’
‘Oui, il l’a repassé.’ / ‘Non, il ne l’a pas repassé.’
The average duration of the individual word regions that were identical across the three question forms (e.g., ‘repassé’, ‘cet’, ‘après-midi’, ‘sa’, ‘chemise’) never differed by more than 30 ms. Paired t-tests indicated that there were no significant differences in word duration across conditions (all p’s > 0.35).
The procedure was the same as in Exp. 2.
Results and Discussion
First, we briefly consider the behavioural data to validate the experimental task. Accuracy was close to ceiling at 96.3% correct responses. Participants were more accurate for negative questions (98.4% correct for ‘yes’ responses and 95.0% for ‘no’ responses) than for positive questions (98.2% correct for ‘yes’ responses and 93.5% for ‘no’ responses). Recall that, for English, we observed the lowest accuracy for ‘yes’ responses to negative questions; for French, accuracy was the highest in this condition. This difference is likely caused by the presence of French ‘si’ as an unambiguous way of referring to a positive state of affairs, whereas in English ‘yes’ is ambiguous between referring to a positive or negative state of affairs.
Response times were also numerically faster for negative questions (1425 ms for ‘yes’ answers and 1582 ms for ‘no’ answers) compared to positive ones (1418 ms for ‘yes’ answers and 1603 ms for ‘no’ answers). However, the difference between positive and negative questions was extremely small. Again, these observations accord with our prior expectations. Hence, we now turn to the eye-tracking data.
Figure 4 shows the proportions of looks to the p and ~p images during the question phase and subsequent 1.5 s gap for each of the three question forms. In the positive condition, there was a stable bias towards the p image in the noun region and in the second half of the gap between question and answer. In the negative condition, there was a small bias to the p image in the gap region.
To test whether the biases were statistically significant, we calculated the natural log ratio of the probability of looks to the p image over looks to the ~p image. We conducted mixed linear regression analyses to determine if the log ratios differed reliably from 0. In these analyses, we included random intercepts and slopes for participants and items. If this analysis did not converge, we stepwise dropped random slopes for items and participants. The results of these analyses are shown in Table 3.
Again, participants initially paid attention to both p and ~p images in both positive and negative conditions. In the positive condition, there was a significant bias towards the p image in the noun region and during the second half of the gap between question and answer. In the negative condition, there was a marginally significant positive bias in the verb region and during the first half of the gap between question and answer.
In order to determine if the log ratio differed across conditions, we conducted mixed linear regression analyses predicting log ratio based on condition (positive or negative), with random intercepts and slopes for participants and items. Again, we stepwise dropped the random slopes if the models failed to converge. The alpha value was corrected for multiple comparisons and set at 0.016. This analysis indicated that, in the noun region, the log ratio in the positive condition was numerically but not significantly higher than in the negative condition (t(79) = 2.3, p = 0.03, model = intercepts only).
The results for positive questions are similar to the results for positive questions in Exps. 1 and 2, showing a mitigated bias for the p image from the noun region onward. The results for negative questions are more difficult to interpret. In particular, there was a marginally significant positive bias already in the verb region. Given that we did not find such an early bias in either of the previous experiments, and given that the effect was only marginally significant, we will not consider this effect in more detail. In addition, there was a marginally significant positive bias in the first half of the gap between question and answer, which is in line with the results of Exps. 1 and 2.
In the next section, we discuss the consequences of these findings for semantic theories of polar questions.