1. Issue
Whether the head of relative clauses (RCs) can reconstruct into the RC at Logical Form (LF) indicates whether there is any syntactic movement of the head (e.g., Brame Reference Brame1968, Schachter Reference Schachter1973, Chomsky Reference Chomsky, Hale and Keyser1993).Footnote 1 There are several commonly used diagnostics to examine the head reconstruction in RCs, one of which is scope assignment (e.g., Bianchi Reference Bianchi1999, Aoun and Li Reference Aoun and Li2003), as illustrated in (1):
(1) I called the [two patients]i [that every doctor will examine ti] tomorrow. (Bianchi Reference Bianchi1999: 46)
This sentence gives rise to two possible interpretations:
(i) I called the same two patients that every doctor will examine tomorrow. (surface scope reading: two > ∀)
(ii) I called the different groups of two patients that every doctor will examine tomorrow. (reconstructed scope reading: ∀ > two)
The existence of the interpretation in (ii), where every scopes over two patients, indicates that the head of the RC two patients can reconstruct into the embedded direct object position at LF. Since reconstruction implies syntactic movement, the reconstruction of the head in (1) further suggests that it is raised from within the RC.
Aoun and Li (Reference Aoun and Li2003) point out that in Mandarin Chinese (hereafter referred to as Chinese) RCs, whether or not the head can reconstruct is determined by its phrasal category. When the head is a noun phrase (NP), it can reconstruct into its base position inside the RC at LF, which is supported by the result of a diagnostic test using anaphor binding (see Schachter Reference Schachter1973):
Under the assumption that anaphors are subject to Condition A (Chomsky Reference Chomsky1981, Reference Chomsky1986), since the Chinese anaphor ziji ‘self’ in (2) is not c-commanded by the most embedded subject meigeren ‘everyone’ on the surface, the head zijide chezi ‘self's car’ must reconstruct into the RC at LF so that the anaphor can be c-commanded by the subject meigeren ‘everyone’. The grammaticality of (2) suggests that the head has raised from an RC internal position.
However, Aoun and Li also observe that such reconstruction is not always available when it comes to scope assignment, as in (3):
They claim that in (3), sanbenshu ‘three books’ must be the same set of three books that everyone will read, which means the head of the RC always scopes over the RC subject meige-ren ‘everyone’. However, if we remove the particle dou, which is a distributor (e.g., Lee Reference Lee1986, Liu Reference Liu1990) and generally co-occurs with a QP (Aoun and Li Reference Aoun and Li1993), the RC subject may scope over the head, as in (4):
In (4), sanbenshu ‘three books’ can be either the same or different three books that everyone will read.Footnote 2 Thus, there is a contrast between (3) and (4): in (3) where there is a dou inside the RC, the only available interpretation is ‘three > ∀’, whereas in (4), where dou is not present, there are two possible interpretations: ‘three > ∀’ and ‘∀ > three’. Aoun and Li (Reference Aoun and Li2003) further note that this contrast cannot be found in nonrelative sentences like (5):
In (5) with/without dou, the interpretation ‘∀ > three’ is available. Thus, in order to account for the contrast between (3) and (4), Aoun and Li argue that if the head of the Chinese RC is a QP having the form [Quantifier+Classifier+Noun], it cannot reconstruct into the RC at LF, which is why the universal QP in the RC subject position cannot scope over the head QP in (3).
But how about (4), where meige-ren ‘everyone’ can scope over sanben-shu ‘three books’? Aoun and Li (Reference Aoun and Li1993, Reference Aoun and Li2003) propose an account which elaborates on Lee (Reference Lee1986): If there is no dou ‘all’ inside the RC, the RC subject meige-ren ‘everyone’ can be raised out of the RC and c-command the head QP sanben-shu ‘three books’ at LF. However, if a dou exists, meige-ren ‘everyone’ cannot be raised due to a domain requirement between dou and its related QP. That is, the universal QP must be in the government domain of dou. Aoun and Li's analysis is summarized in (6):
(6)
a. The head QP in Chinese RCs cannot simply reconstruct into the RC.
b. When there is no dou inside the RC, a universal QP in the RC subject position can be raised out of the RC at LF and have scope over the head QP.
c. When there is dou inside the RC, a universal QP in the RC subject position cannot be raised out of the RC because such QP has to be within the domain of dou.
In short, if the head of a Chinese RC is a QP, it cannot reconstruct into the RC. However, if it is an NP, reconstruction can occur. This constraint on QP-headed RCs is further complicated by the presence of dou ‘all’. However, Aoun and Li's (Reference Aoun and Li2003) judgment on (3) and (4) has not been rigorously tested by experiments.
This study used a sentence-picture matching Truth Value Judgment (TVJ) task (Crain and Thornton Reference Crain and Thornton1998) to investigate the following question: When dou ‘all’ occurs inside a QP-headed Chinese RC that has a universal QP in its subject position, does the head QP always have scope over the universal QP? The experimental results showed that the universal QP may scope over the head QP and that the contrast between (3) and (4) claimed by Aoun and Li (Reference Aoun and Li2003) does not seem to exist.
2. Experiment
2.1 Participants
A total of 16 adult native speakers of Chinese participated. They were all university students in China, with ages ranging from 19 to 25. They were all born and raised in China and none of them had lived outside China at the time of the experiment. They were paid for their participation.
2.2 Design and materials
A sentence-picture matching TVJ task was created, where seven characters from the Chinese novel Journey to the West were used.Footnote 3 There were four students, Monkey (7a), Monk (7b), Pigsy (7c), and Sandy (7d), and their teachers, Sakyamuni (7e), Maitreya (7f) and Bodhisattva (7g), shown below:
Each experimental item begins with a story. Here is one example: One day, the four students made their own chocolate and they put photos of their faces on it, as illustrated in (8):
The story continues, as follows. The three teachers tasted different students’ chocolate: Sakyamuni tasted Monkey's and Monk's chocolate, Maitreya tasted Monk's and Pigsy's chocolate, and Bodhisattva tasted Pigsy's and Sandy's chocolate. Then a dog appears, as in (9), and says either (10) with dou or (11) without dou (presented in aural form only):
The story for each experimental item was presented using both pictures and written sentences in Microsoft PowerPoint. All target sentences were pre-recorded by a native Chinese speaker with natural intonation. The target sentences varied by whether the particle dou ‘all’ occurs in the target sentence, which results in two conditions: (i) dou RC (DRC) condition and (ii) plain RC (PRC) condition, as exemplified in (10) and (11) respectively. The participants were then asked to judge whether the sentence and the picture matched by saying dui ‘yes’ or cuo ‘no’. Note that in both (10) and (11), the surface scope reading entails the reconstructed scope reading. Therefore, the only way we can test the availability of the reconstructed scope reading is to create a context where the dog's utterance is true under the reconstructed scope reading but false under the surface scope reading.Footnote 4 The story in (9) provides only a union reading context.Footnote 5 That is, in order for (10) and (11) to match (9), the reconstructed scope reading must be true, where the universal QP meige laoshi ‘every teacher’ in the RC subject position scopes over the head QP liangfen chaokeli ‘two portions of chocolate’. The surface scope reading, on the other hand, must be false in the context of (9). Thus, if both (10) and (11) allow the reconstructed scope reading, we expect the participants to accept both types of sentences. However, if (10) only allows the surface scope reading while (11) allows both readings, as claimed by Aoun and Li (Reference Aoun and Li2003), we expect the participants to reject (10) but accept (11).
Participants were presented with one trial at a time on a computer screen. A total of 16 items of different lexicalizations were created, each of which had two conditions, resulting in 32 tokens. The target trials were further distributed to two lists so that each list contained only one condition from the same lexicalization. Thus, there were 16 critical stimuli in each list, and each condition had eight stimuli. Two blocks were created for each list and each block involved only one critical condition. Specifically, there were eight DRC trials in Block I and eight PRC trials in Block II of each list. Trials of the same condition were grouped into a single block because if participants see both conditions in the same list, their judgment on DRC trials may be affected by their judgment on PRC trials. Note that since the stories embedded in DRC and PRC trials were very similar and the only apparent difference between them is the particle dou in the target sentence, participants might easily detect the difference and respond accordingly.Footnote 6
For each of the 16 experimental items, apart from the critical sentence, there were two filler sentences included, which were categorized as Type I filler and Type II filler respectively. The Type I fillers were used to monitor whether participants had fully understood how to complete the experiment. The Type I filler for the scenario in (9) is shown in (12):
Participants were expected to say ‘no’ to the trial involving the sentence in (12) and the picture in (9), since the picture shows that Sakyamuni did taste Monkey's chocolate. There were eight ‘yes’ and eight ‘no’ trials in each list. Meanwhile, the Type II fillers were used to monitor whether participants had paid attention to the presence of dou in the experimental sentences, as our DRC and PRC trials differed only in the involvement of dou. The Type II filler for the scenario in (9) is in (13), which is with/without dou:
Without dou, (13) allows both collective and distributive readings, but if dou occurs, it only allows the distributive reading. According to Lin (Reference Lin1998), if a sentence allows both collective and distributive readings, adding the particle dou to that sentence would block the collective reading. Thus, our participants were expected to say ‘yes’ to the trial without dou and ‘no’ to the trial with dou. There were eight ‘yes’ and eight ‘no’ trials in each list.
To sum up, each experimental item had a story, which included a critical trial, a Type I filler and a Type II filler. The order of the three trials within each experimental item was pseudorandomized and the very first trial in each list was a filler. The ‘yes’ and ‘no’ fillers were evenly distributed across items and the order of all experimental items was randomized.
2.3 Procedure
The experiment was conducted on a one-to-one basis using VooV Meeting, a platform similar to Zoom. At the beginning of the experiment, each participant completed a background information survey, which included (i) name; (ii) hometown; (iii) native language; (iv) experience of living abroad.
To ensure that the participants understood how to do the TVJ task, several examples and practice trials were presented before the target trials. Two of the sample trials are below: one trial of them included (14) and (15a) while the other trial included (14) and (15b):
In (14), the pronoun ta ‘he’ can refer to either the subject Sunwukong ‘Monkey’ or Shaheshang ‘Sandy’. Participants first saw the picture (15a) and then listened to (14). If they said ‘no’, they would be asked to think about whether there is any possibility of saying (14) in the context of (15a). All the participants who had initially said ‘no’ successfully switched to ‘yes’ after reconsideration. Then the trial involving (15b) and (14) was presented and all participants said ‘yes’. After going through this pair of examples, participants were explicitly told that if a given sentence had two possible interpretations, the trial should be accepted so long as there was one interpretation that matched the picture. This procedure was used to guide participants towards making their judgment based on acceptability rather than preference. As pointed out by White et al. (Reference White, Bruhn-Garavito, Kawasaki, Pater and Prévost1997), the effects of preferences for certain interpretations over others cannot be fully eliminated in TVJ tasks, and we can only minimize them with different methods. The following section examines each individual participant's data and whether they made a distinction between DRCs and PRCs in the union reading context.
2.4 Results
Recall that the Type II fillers were used to monitor whether the participants paid enough attention to the presence of dou in experimental sentences. Since there were eight ‘yes’ and eight ‘no’ trials, based on the binomial cumulative distribution, participants were expected to accept seven or more ‘yes’ trials and reject seven or more ‘no’ trials.Footnote 7 This applied to the Type I fillers as well: each participant should have accepted seven or more ‘yes’ trials and rejected seven or more ‘no’ trials. Table 1 summarizes the 16 participants’ mean proportions of ‘yes’ answers for the fillers:
Summary of the participants’ mean proportions of ‘yes’ answers for the fillers
The individual data reveals that all participants accepted seven or more ‘yes’ trials and rejected seven or more ‘no’ trials for Type I fillers, and that 15 of them did so for Type II fillers. One participant wrongly accepted four ‘no’ items, and so their data was removed. All of the 15 participants who passed the initial screening accepted eight DRC items and eight PRC items, which strongly indicates that the universal QP can scope over the head QP in the target Chinese RC, regardless of whether the RC involves dou or not.Footnote 8 This finding goes against Aoun and Li's (Reference Aoun and Li2003) statement that when a particle dou occurs inside a QP-headed RC with a universal QP in the subject position, the universal QP cannot scope over the head QP.
3. Discussion and conclusion
The experimental results suggest that in Chinese, the universal QP in the subject position of QP-headed RCs can scope over the head QP, irrespective of the presence of dou. Thus, the head QP should freely reconstruct into the RC at LF, which in turn has an important implication for the availability of head reconstruction in Chinese RCs.
As we have seen, Aoun and Li (Reference Aoun and Li2003) note that head reconstruction is inconsistent in Chinese RCs, summarized below:
(16)
a. Reconstruction is possible for binding relations involving anaphors, bound pronouns, and so on, in the Head.
b. Reconstruction is not possible for structures involving a Head QP interacting with another QP inside an RC for scope interpretations.
(Aoun and Li Reference Aoun and Li2003: 139)
Following (16), Aoun and Li conclude that it is the NP/QP distinction that leads to the conflicting reconstruction effects: only NP heads can reconstruct. Nevertheless, the data from Chinese participants reported in this study strongly suggest that such a stipulation is unnecessary. Instead, head reconstruction should always be possible for Chinese RCs, irrespective of the phrasal category of the head.
Moreover, according to Lee (Reference Lee1986) and Aoun and Li (Reference Aoun and Li1993, Reference Aoun and Li2003), the union reading in PRCs such as (4) and (11) can be explained by subject raising: The universal QP in the RC subject position can be raised out of the RC to a higher clause and take scope over the head QP at LF. The subject raising was attributed to the fact that Chinese lacks Agreement and the embedded IP cannot be a governing category for the trace of the raised QP to be properly bound. In contrast, when there is a dou inside the RC, such as (3) and (10), the universal QP cannot be raised out of the RC due to the domain requirement of dou. Aoun and Li (Reference Aoun and Li2003) provide a single piece of evidence arguing for the subject raising analysis: If a universal QP is located in a doubly embedded subject position, it cannot be raised out of the RC to have scope over the head QP, as in (17):
In (17), the universal QP meige-ren ‘everyone’ is located in the subject position of a doubly-embedded RC. According to Aoun and Li, a union reading is disallowed because the universal QP can only be raised to its immediate higher clause and adjoined to the VP headed by zhengli ‘arrange’. However, Su and Chen (Reference Su and Chen2022) ran an experiment to test this and found that a union reading is indeed possible in sentences like (17). That is, even when the universal QP is embedded in a doubly-embedded RC, it still can have scope over the head QP, which suggests that the union reading in PRCs should be explained by head reconstruction rather than subject raising.
To conclude, this sentence-picture matching TVJ task showed that in Chinese QP-headed RCs with a universal QP in the subject position, the universal QP can scope over the head QP, regardless of whether the RC involves dou or not. Thus, we do not need to stipulate that there exists an NP/QP distinction regarding head reconstruction in Chinese RCs. Rather, it is possible for both NP and QP heads to reconstruct into the RC at LF, predicted by the head-raising analysis of RCs (e.g. Brame Reference Brame1968, Schachter Reference Schachter1973, Kayne Reference Kayne1994).
