Domain restriction in child Mandarin: Implications for quantifier spreading

1.1.1 Quantifier spreading and the event quantification account

Studies on the acquisition of quantifiers have uncovered a puzzling phenomenon among 3- to 10-year-old children. For example, Philip (1995) presented adults and children with a picture depicting three pigs and four apples, where each pig is eating one apple, leaving one apple uneaten (in the picture there is also a dog eating a bone). When seeing this picture, almost all the adults said that (1) is a correct description of the scenario, because for adults the domain of the universal quantifier every includes only the subject NP pig, not the object NP apple.^[1]

On the other hand, children between four and five years old judged (1) to be false. The children, pointing to the extra apple (which we will call the extra object), typically provided the following reason to justify their judgment: there is no pig eating the extra apple. In other words, they seemed to require an exhaustive pairing between the pigs and the apples.

What has led to such a non-adult response? Philip (1995) proposes that the exhaustive pairing is a result of a symmetrical interpretation as shown in (2),^[2] where the quantifier every “spreads” its domain to the object NP. The exhaustive reading is thus also called quantifier spreading, which is also coined as symmetrical responses (cf. Inhelder and Piaget 1964; Roeper and Mattei 1975 for earlier studies on the topic).

The quantifier spreading problem leads to a question that interests many researchers: Are 4- to 5-year-old (and even older) children competent with universal quantification/distributive computation? According to Philip (1995), quantifier spreading is driven by a non-adult semantic representation of the universal quantifier. Philip argues that children may misunderstand every to be a universal quantifier that takes events, rather than individual objects, as its domain, as shown in (3).

In other words, for adults, every quantifies over only the NP it syntactically merges to, taking that NP and only that NP as its domain, but for children, every may take the whole event as its domain.

1.1.2 Problems of the event quantification account

Researchers have raised several concerns regarding Philip’s (1995) account. First, the dual-status of every, that is, every can both quantify over events or sets of individuals, causes a learnability problem. Since children still provide adult-like answers in addition to the symmetrical responses, we must assume that children can interpret every either as a determiner quantifier or as an adverbial event quantifier like always. Therefore, adults’ productions will always be consistent with one of the assumed interpretations that children assign, and so children will need to unlearn the event quantification interpretation (see examples 2 and 3 above) in the absence of negative evidence (Crain et al. 1996; Philip 2011).^[3] We will return to this acquisition issue later in Section 3. Second, as pointed out by Crain et al. (1996), an important source of children’s symmetrical responses is the flawed experimental design, i.e., the presence of an extra object, e.g., the fourth apple in (1), in the pictures used in the experiments, which might give a hint that this extra object was relevant to the truth value judgment of the test sentence. In other words, a child who takes the extra object as pragmatically relevant for the judgment of the truth value of (1) will conclude that the correct and expected answer is “no.” In other words, a child may be attracted by the extra object, reasoning that (1) should be judged by taking the extra, uneaten apple into consideration. In this case, the child will simply choose the pragmatically felicitous response and violate a syntactic principle of quantifier domain restriction that is in conflict with an overriding pragmatic principle.

Crain et al. (1996) argue that Philip’s simple ‘yes/no’ task is a flawed experimental design because no alternative to assertions like (1) was provided. If (4a) and (4b) were both provided as possible outcomes of the story, Crain et al. (1996) predict that children would not be attracted by the extra objects. This is because if (4b) is instead the actual outcome, (4a) would be false. Therefore, the assertion in (4a) is in doubt at some point in the story. This satisfies the condition of plausible dissent (p. 116; see also Freeman et al. [1982]).

In a series of experiments, Crain et al. (1996) argue that such a design flaw can be ameliorated by introducing both (4a) and (4b) as possible outcomes at certain point of the trail. Experiments conducted by Crain et al. (1996) validate this proposal, confirming that children do not have problems with universal quantification and distributive computation.

We agree that the pragmatic salience of the extra object might have attracted the subjects’ attention in Philip’s (1995) experiments, and in our experiments we will avoid such a design, but we argue that the effects of extra objects shed light on the nature of the problem of quantifier spreading. In addition, even though it is true that under certain conditions children’s symmetrical responses can be eliminated, we still lack a principled explanation for why, specifically under the single-extra-object condition, such a high percentage of non-adult symmetrical responses were observed (Minai et al. 2012; Philip 2004).

1.1.3 Salience effects and domain restriction problems

A number of other explanations for children’s symmetrical responses can be categorized as pure syntactic/semantic deficiency, or syntax-to-semantics mapping deficiency approaches (Brooks and Sekerina 2006; Roeper and de Villiers 1993; Roeper et al. 2011, 2005), in keeping with Philip (1995), assuming children’s symmetrical responses are due to their non-adult knowledge status regarding universal quantification or distributive computation. Still others advocate for pure pragmatic or cognitive deficiency approaches (de Koster et al. 2018; Gouro et al. 2001; Gualmini et al. 2003; Minai et al. 2012; Rakhlin 2007), which Crain et al. (1996) also argue for, suggesting that children do not have problems with universal quantification, but rather the observed non-adult behavior in universal quantification is caused by pragmatic factors inadvertently induced by the experimental design. Another popular string of proposals are combinatory approaches, the weak quantification hypothesis specifically, resorting to both syntactic/semantic and pragmatic factors (Drozd 2001, 2004; Drozd and van Loosbroek 2006; Drozd et al. 2017; Geurts 2003), i.e., children’s weak quantifier representation of every allows pragmatic factors to have an important influence on the interpretation of every.

Our review below will focus on a specific account of quantifier spreading, which leads to our study. Note that all the three strings of proposals mentioned above must explain why the domain of the universal quantifier can be extended to another NP. Both the syntactic/semantic deficiency approaches and the weak quantification approaches, for example, argue that children have a non-adult interpretation of universal quantifiers, whereas the pragmatic approaches do not have a principled account. We agree with the pragmatic/cognitive approaches that children may have no problems with universal quantification, and evidence from Mandarin Chinese will be provided below. However, the main difficulty for this approach is that a principled explanation, besides the flawed experimental design, should be provided to explain why quantifier spreading occurs (e.g., in extra-object conditions). If quantifier domain extension is altogether not allowed, pragmatic factors are quite unlikely to even have an opportunity to sneak in.

A specific pragmatic factor, salience effects, will be employed below to demonstrate that children must have non-adult domain restriction. Various recent studies converge on the conclusion that the extra object in studies such as Philip (1995) is significantly correlated with the emergence of quantifier spreading. We argue that extra objects can lead to quantifier spreading because the salience effects associated with the extra objects may affect the selection of the quantifier domain for strong quantifiers. Such salience effects are reported extensively in the literature.

On the one hand, increasing the salience of the target quantifier domain (the NP that every merges to) in adult grammar reduces the proportion of symmetrical responses significantly (Crain et al. 1996; Drozd and van Loosbroek 2006; Hollebrandse 2004).^[4] If the target quantifier domain is brought to children’s attention or is introduced as a discourse topic before the presentation of the test sentence, the target domain will be more likely correctly selected. For example, Freeman et al. (1982) show that children’s responses may be affected by certain variation in the presentation of experimental stimuli. (see also Drozd [2001]). Freeman et al. find that, if a story is about cows, then children paid more attention to cows, and 89% of the participants required an exhaustive interpretation on cows; but when the story was about cowsheds that were burning down, 77% of the same children adopted a strategy that requires an exhaustive interpretation of cowsheds in the story context. In addition, Hollebrandse (2004) clearly shows that when the subject NP (the domain of the quantifier) is also the discourse topic, the symmetrical responses elicited by pictures with an extra object decreases significantly.

On the other hand, decreasing the salience of the extra object reduces symmetrical responses. Kiss and Zétényi (2017) replicated quantifier spreading in Hungarian five-year-olds with iconic drawings typical of previous studies on quantifier spreading. However, when replacing these iconic drawings with photos taken in a natural environment that are rich in accidental details, the occurrence of quantifier spreading was radically reduced. Kiss and Zétényi (2017) argue that in traditional pictures with one extra object, the extra object is taken as relevant to the judgment task, given Csibra and Gergely’s (2009) Natural Pedagogy theory (based on Sperber and Wilson’s [1995] Relevance Theory). An important effect alongside the distracting and noisy background in the pictures is that the extra object becomes less prominent and less relevant to the judgment of the test sentences. Therefore, it could be that the extra objects are less salient in the photos than in the iconic drawings (Gordon 1998), which might have led to a reduction of the symmetrical responses.

Similarly, Gouro et al. (2001) decreased the salience of the extra object with two design features. In one condition, Gouro et al. employed figures where the agents are of different types (e.g., different Pokemons) and the objects are of different colors (e.g., different colors of ponies that the Pokemons are riding). Therefore, both the agents and the objects are quite different from each other, and there are no extra objects that have a unique exceptional property. Gouro et al. argue that such a setting reduces the uniqueness and salience of the extra object, compared to Philips’ (1995) classical extra-object pictures, where the extra object stands out as a unique exception. In another condition, Gouro et al. included a context before the presentation of the traditional single extra object pictures, and in that context, they made the extra object less relevant to the truth value of the test sentence. For example, while the test sentence is about cats riding ponies, the context given before the presentation of the test sentence is not about riding but about whether the cats should get close to the ponies. In both conditions, the percentage of symmetrical responses dropped significantly.

In addition, Freeman (1985) shows that when the extra object is paired by an irrelevant, unmentioned agent, non-adult symmetrical responses are also significantly reduced, suggesting that the presence of this unmentioned agent lessens the salience of the extra object for the children (Drozd 2001), and therefore leads to more adult-like responses.

Lastly, increasing the number of the extra objects also decreases the salience of each of the extra objects and thus reduces quantifier spreading (Minai et al. 2012; Sugisaki and Isobe 2001). Sugisaki and Isobe (2001) included 6–7 extra objects in the testing pictures, and Minai et al. (2012) used three extra objects. Both studies find that with multiple extra objects, the percentage of symmetrical responses decreases significantly. Again, these are cases where the extra objects lose their uniqueness and become less salient than those in traditional single-extra-object pictures. Furthermore, Minai et al. (2012) find that when pictures depicting multiple extra objects were presented before pictures with a single extra object, the percentage of symmetrical responses was lower than when the single-extra-object pictures were presented first, indicating that some non-syntactic/semantic factors must have an effect on quantifier spreading. Minai et al.’s eye movement data also show that the mean proportion of eye fixation on the extra object(s) in the single-extra-object condition is much higher than that in the multiple-extra-object condition. Minai et al. (2012) thus suggest that with multiple extra objects, the extra objects lose their “uniqueness” status as in the single-extra-object condition, decreasing the salience of the extra object.

The multiple-extra-objects design may also have contributed to the decrease in the symmetrical response observed in Crain et al. (1996). The scenarios in Crain et al. (1996) typically involve more extra objects than those in Philip (1995). For instance, there are three extra objects in Experiment 4, and 2 extra objects (in addition to five other extra objects not mentioned in the test sentences) in Experiments 2, 3 and 5.

Of course, a formal theory of salience (in the visual domain as well as the linguistic domain) is needed in order to make a decisive evaluation. However, the evidence available seems to suggest that the salience of the extra object and the target domain of the quantifier has an important impact on the percentage of symmetrical responses obtained in previous studies. This leads to the question why the salience effects associated with extra objects directly influence the emergence of quantifier spreading. We argue below that this is because of children’s uncertainty of domain restriction in universal quantification.

2.2.1 Methodology

2.2.1.3 Materials

Our first experiment included 12 target trials and 12 fillers. A typical test sentence is given in (16), as we have already discussed in (8), in which there are two NPs to the left of dou and c-commanding dou: wugui ‘tortoise’ and laoying ‘eagle’.^[15]

(16)

Wugui	zai	laoying-pangbian	dou	zhong-le	lanhua.
tortoise	p	eagle-loc	all	plant-asp	orchid
‘The tortoise(s) at the eagle(s) side all planted orchid(s).’

Either NP can be the domain of dou. The test sentence was presented following a scenario depicted either as in Figure 2 under the Subject Condition (six items; corresponding to the reading in which dou quantifies over the subject NP wugui ‘tortoise’) or as in Figure 3 under the Preposition Object Condition (six items; corresponding to the reading in which dou distributes over the object of the preposition, laoying ‘eagle’). The size of characters (drawings of either human or animals) was kept as similar to each other as possible. The position of the objects (the orchids in Figures 2 and 3) is counterbalanced across items.

Figure 2:

Subject condition.

Figure 3:

Preposition object condition.

The story associated with Figure 2 is as in (17).

(17)

Here are two tortoises and two eagles. The tortoises want to plant orchids beside the eagles. Look, at the side of this eagle (pointing to one of the eagles) there is an orchid. Who planted it? Is it this tortoise or that tortoise (pointing to both tortoises)? Yes, it is the tortoise on the left that planted it. What about the other orchid? Hmm, it is also planted by the tortoise on the left. You see, the tortoise on the right (pointing to the tortoise on the right side) didn’t plant any orchid.

A story with the same pattern was presented when Figure 3 was shown to the participants.

Note that when presenting the test sentences, we put the prosodic stress on dou, and left the two NPs un-emphasized, in order to avoid an influence from the emphasis pattern on domain selection.

Fillers were created to ascertain if children pay attention to the test. The fillers were designed such that half would be answered “yes,” and the other half “no.” The results of the fillers can shed light on the participants’ understanding of dou-quantification over the subject NP or the preposition object. The fillers were of two types; either only the subject NP is available to be quantified by dou (18a), or only the NP complement of the PP is available (18b).

(18)

a.

Wugui	dou	zhong-le	lanhua.
tortoise	all	plant-asp	orchid
‘The tortoises all planted an orchid.’

b.

Laoying-pangbian	dou	zhong-le	lanhua.
eagle-loc	all	plant-asp	orchid
‘Besides all the eagle(s) there is an orchid planted.’

2.2.2 Results

The results of the 36 adults and 24 children (12 for the Subject Condition and 12 for the Preposition Object Condition) were included in the data analysis. The results are presented in Figure 4.

Figure 4:

Adults’ and children’s acceptance rates of the test sentences in Experiment 1 (error bars are standard errors; the dots represent the average acceptance rates of each participant).

As shown in Figure 4, adults prevailingly accepted the test sentences in the Preposition Object Condition at a rate of 79%, whereas they accepted the test sentences in the Subject Condition at a rate of only 23%. Note that these numbers are from two separate conditions and thus they both suggest that adults may prefer to take the preposition object as the domain of dou. A Pearson’s Chi-squared test reveals that the difference between the acceptance rates under the two conditions reaches statistical significance at α = 0.05, with X ² (1, 432) = 133.36, p < 0.001. The statistical test confirms that adults have a preference to choose the NP object of the preposition instead of the subject NP as dou’s domain. However, children do not have such a preference, since the acceptance rates of the test sentences under both conditions are 42%.^[16]

A closer examination of the individual results (the dots in Figure 4) underpins the idea that adults have a strong preference to select the object of the preposition as the domain of dou, whereas children do not have such a preference; on the contrary, there seem to be more children that were reluctant to take the preposition object than the subject as the domain. Five children rejected all the test sentences under the Preposition Object Condition, whereas only two children rejected all test sentences under the Subject Condition. In addition, there was much variation across individuals, which suggests that their selection of domain may be based on a probabilistic mechanism, which will be discussed in detail in Section 3.

2.2.3 Discussion

Our results confirm Cheng’s (1995) proposal regarding adults’ grammar that either the subject NP or the object of the preposition can be the domain of dou. However, it is not predicted that the object of the preposition is taken as the domain most of the time. These results suggest that dou is not a strict unselective universal quantifier (Lee 1986; Lewis 1975) that must quantify over all free variables in its scope. Indeed, if dou is a strict unselective universal quantifier, then we would expect dou to quantify over both bare nouns simultaneously. The adults would therefore have rejected the test sentences in both conditions. On the contrary, it turned out that adults could accept the test sentences in either the Subject Condition or the Preposition Object Condition, although the object of the preposition is the preferred domain (see Ke et al. [2018] for another piece of evidence showing that dou cannot be a strict unselective universal quantifier).

A similar conclusion can be drawn for the child participants: children do not require dou to quantify over all NPs in its scope. If children require dou to quantify over both NPs in the test sentences, they would always reject the test sentences under both conditions, which is rarely observed, as indicated by the individual data especially under the Subject Condition in Figure 4; namely, only two child participants completely rejected the test sentences.

The question immediately arises as to why adults prefer to take the object of the preposition rather than the subject as the domain of dou. One important reason may be that the object of the preposition is linearly and structurally closer to dou than the subject. As we have seen in the literature review, domain selection is influenced by various salience effects. The object of the preposition may thus be more salient than the subject NP because of being linearly and structurally closer to dou; that is, the path between the preposition object NP node and dou is shorter than the path between the subject NP and dou (Cheng 1995; Lee 1986; Pesetsky 1982). However, this explanation has to assume that children, on the other hand, are not sensitive to structural locality.

Another possible explanation is that adults may employ pragmatic reasoning based on the distribution of cases of dou-quantification. That is, when dou is located immediately after the subject NP as in (19), it unambiguously quantifies over the subject NP.

Therefore, adults may perceive dou’s being located immediately after the PP as a pragmatic clue implying that dou is meant to quantify over the preposition object. In other words, the adult participants may reason: if the speaker’s intention were to have dou quantify over the subject NP specifically, why did the speaker use an ambiguous construction instead of the unambiguous one as exemplified in (19)? Child participants, on the other hand, may lack this kind of pragmatic reasoning which is based on the linguistic distribution of competing constructions.

Children accepted the test sentences under both conditions 42% of the time. These results can be interpreted in two different ways. One interpretation may be that children answer the questions randomly, and perhaps do not understand the test sentences at all. However, this interpretation seems quite unlikely. We want to highlight below several reasons why we do not believe that the children’s responses were based on random guesses, with a lack of understanding the meaning of the universal quantifier dou or even that of the whole test sentences. On the contrary, we believe that the children might “randomly” select an NP from the two NP candidates with a proper understanding of both the meaning of dou and that of the whole test sentences.

Importantly, the children provided adult-like justifications to support their “no” responses to test sentences. For example, when providing a reason why the test sentence (16) is false with regard to the story under the Preposition Object Condition, the children typically said “No, just one of the two tortoises planted an orchid,” which is a clear indication that they really understood the story and the test sentence and did not respond by guessing. If they did not understand the meaning of dou or the meaning of the whole test sentences, such right-to-the-point answers are surprising.

In addition, the children displayed adult-like competence in filler items such as those in (18) with a correct rate of 95%. These filler sentences also involve dou-quantification, and therefore the adult-like responses to these sentences strongly indicate that the children understood the meaning of dou and also took dou into consideration when judging the truth value of these sentences.

Furthermore, we only included in analysis the children who understood the universal force carried by dou in our pretest (Section 2.1). That is, those children who did not distinguish a sentence with dou from one without dou have already been excluded from the analysis.

Lastly, various previous studies on dou have showed that children understand the semantics of dou-quantification (Hsieh 2008; Lee 1986; Zhou and Crain 2011), as we have reviewed in Section 1.4.

An alternative interpretation, which we believe is much more plausible, is that it was not easy for child participants to select a proper NP as the domain of dou in the test sentences. These two NPs are syntactically close to each other, and since both of them are available to be the domain of dou, children are predicted to experience much interference when computing the domain of dou. Such competition effect is discussed in Kurtzman and MacDonald (1993) and Paterson et al. (2008). Basically, if a sentence is ambiguous, one of the interpretations would decrease the acceptability degree of the other. According to the cue-based retrieval model (Van Dyke and McElree 2006; Lewis and Vasishth 2005; Lewis et al. 2006), competitors sharing features with the target will induce similarity-based interference in memory retrieval. In the case of our test sentences, when dou is encountered, the parser needs to retrieve its domain. Either one of the NPs can be retrieved because they share almost all features, giving rise to severe interference effects. In other words, when children consider one of the NPs as the domain, another NP is also considered a competing domain. Consequently, it appears that children randomly picked one of the NPs as the domain. Different from children, adults may be affected by the similarity-based interference to a less extent, probably because adults’ memory retrieval is less noisy. This conjuncture is supported by a recent study showing that children were temporarily more distracted than adults when multiple retrieval cues supported a prominent competitor antecedent in the processing of reflexive binding (Clackson et al. 2011).

In order to further investigate children’s strategy in domain restriction, another experiment was conducted in which the two plural NPs, the subject and the object, are separated, with the subject placed inside the scope of dou and the object outside. We expect, when the two candidates are separated, the similarity-based interference discussed above can be reduced, and so we may see a cleaner picture.

2.3.1 Methodology

2.3.1.3 Materials

Each subject received one pretest, four test items and four fillers. Pretests used a sentence with dou. A typical example of the test sentences, as well as the pictures accompanying the stories, can be found in Figure 5 and sentence (20).^[17]

Figure 5:

A typical context of Experiment 2.

(20)

Xiaodongwu-men	dou	rengdiao-le	dianshi,	chuang,	he	bingxiang.
animal-plu	all	chucked-out-asp	TV	bed	and	refrigerator
‘The animals all chucked (a) TV, bed and refrigerator out.’

The test sentence (20) was presented following a story corresponding to Figure 5, in which one animal (the Tweety Bird) chucked out a TV and a bed, and the other (the dinosaur) a TV and a refrigerator. As we have mentioned, in adult grammar, dou must distribute over only the NPs that c-command it. In (20), dou should distribute over only the subject NP xiaodongwu-men ‘animals’ which c-commands dou. This leads to the reading in (12a), repeated below in (21). According to the story presented with Figure 5, (21) is false, because the only thing that both animals chucked out is the TVs.

(21)

∀x [x ∈ [[animals]] → x chucks out (a) [[TV, bed, and refrigerator]]]

The scenario where both the animals chucked out the same kind of things is to provide a strong reason for participants who have acquired the adult grammar of dou-quantification to reject the test sentences. This is because for participants who can access the adult grammar and thus take only the subject NP as dou’s domain, the correct statement must instead be ‘The animals all chucked a TV.’ Such a design satisfies the condition of plausible dissent, which is a crucial design feature of the Truth Value Judgment Task (Crain and Thornton 1998).

On the other hand, the conjunct object NP dianshi, chuang, he bingxiang ‘TV, bed, and refrigerator,’ which does not c-command dou, is impossible to be the domain of dou in the adult grammar. This is a good case for us to test whether children can extend the domain of dou to an NP that does not c-command dou, different from adults. The prediction is that if dou distributes over the conjunct object NP in (20), the derived interpretation will be “For the TV, bed, and refrigerator, they are all chucked out by the animals,” as is translated in (12b), repeated below in (22). According to the story presented with Figure 5, (22) is true.

(22)

∀x [x ∈ [[TV-bed-and-refrigerator]] → animals chuck-out x]

In addition to the test trials, each subject saw four fillers. Fillers were designed to verify whether children paid attention to the tasks by asking them simple questions about the stories. Furthermore, we used fillers to diversify the statement patterns, in order to conceal the purpose and the inherent pattern of the target trials. For instance, given the context such as Figure 5, a filler would be Xiaoniao rengdiao-le dianshi, dui haishi budui? ‘The Tweety Bird chucked out a TV, true or false?’ Children were expected to give a Yes–No answer.

2.3.2 Results

All the subjects, including adults and children, accepted or rejected the filler items at an accuracy rate of 100%, so all the data were included in the analysis. The findings shown in Figure 6 are that children accepted the test sentences 79% of the time and rejected them only 21% of the time.

Figure 6:

Acceptance rates of the test sentences of children and adults in Experiment 2 (error bars are standard errors; the dots represent the average acceptance rates of each participant).

The reasons provided by children show that they rejected the test sentences because they assigned the subject NP as the domain of dou. In contrast, adults overwhelmingly rejected the test sentence by 93%. We performed a Pearson’s Chi-squared test in R (R Core Team 2014), which shows that children accepted the test sentences much more often than adults, with X ² (1, 121) = 62.23, p < 0.001.

2.3.3 Discussion

The fact that almost all of the adults rejected the target sentences shows their firm compliance with the c-command requirement as a domain restriction for dou-quantification. Adults were reluctant to select the domain from the NPs outside the scope of dou. On the other hand, four- to five-year-old children showed great uncertainty of domain restriction concerning distributive computation, because they can take an NP that is outside of the scope of dou as the domain of dou.^[18] Individual results further reveal that 9 out of the 12 children accepted all test items, whereas only 2 of the 12 rejected all test items, probably signaling individual differences in the acquisition of the domain restriction associated with dou-quantification.

These results support the idea that children are more uncertain than adults about the domain restriction associated with the distributive operator dou. When more than one bare NP is present, it is possible for children to take either of them to be the domain of dou, even in cases when an NP is outside the syntactic scope of dou.

These results also shed light on the problem raised in the first experiment. It was the syntactic closeness and similarities between the two NPs that might have caused severe interference effects in domain selection to children, resulting in difficulties to identify a proper domain. Therefore, when the two NPs were separated and these interference effects were reduced, children could easily select the NP that is outside of the scope of dou as its domain. There were two children overwhelmingly rejected the test sentences, suggesting that some children considered it obligatory to take the c-commanding subject NP as the domain. Therefore, the acceptance and rejection data together indicate that children can take either the subject or the object NP as the domain of dou.

Given that children are rather uncertain about the domain restriction of dou-quantification, there must be reasons why they biased toward the object NP, as we did not detect such a preference in Experiment 1. A readily available potential factor is that the object NP is a complex NP, which may be more salient than the subject NP, resonating with the salience effects observed in previous studies. This point can be tested further with an experiment where the number of the NPs inside the conjunct NP is reduced, to see if the acceptance rate of the object complex NP would decrease. We leave this to future studies.