1. Wh-Optionality Across Languages
Across languages, wh-questions are expressed by two main strategies. One involves having the wh-element at the sentence-initial position, as shown in English in (1). The classic syntactic approach to wh-fronting involves wh-movement of the wh-element to the specifier position of the CP headed by a null complementizer head with a question feature [+Q] shown in (2). In this paper, I will use ‘wh-fronting’ as a theory-neutral term to refer to questions like (1) and use ‘wh-movement’ as an analysis for wh-fronting questions.
As the other strategy, the wh-element stays in its base-generated position, hereinafter labeled as wh-in situ. Languages including Mandarin Chinese and some varieties of English, for example, Colloquial Singapore English (CSE), use this strategy as shown in (3) where the wh-element stays in the object position. Note that CSE optionally marks tense and includes a set of sentence-final particles like ah, as shown in (3b).
wh-in-situ questions have received different analyses. The covert/LF movement analysis (Huang Reference Huang1982) argues that in wh-in-situ questions, the same wh-movement occurs in a covert component of grammar, usually identified as the logical form (LF). Thus, the structure for sentences like (3a) is identical to the one in (2), except that the wh-element is pronounced at its base-generated position, that is, the object of the verb. On the other hand, the unselective binding analysis (Pesetsky Reference Pesetsky, Reuland and ter Meulen1987, Tsai Reference Tsai1994 among others) proposes that wh-in-situ questions do not involve wh-movement at all. Instead, the C
$ {}_{\left[+Q\right]} $
head binds the wh-element at its base-generated position, as is shown in (4). The binding relation is indicated by the index i. See Lu et al. (Reference Lu, Thompson and Yoshida2020), Tian et al. (Reference Tian, Park and Yang2022) for a recent discussion of the two analyses based on experimental evidence. Putting their differences aside, both analyses of wh-in-situ questions utilize the syntactic feature [+Q] to derive the question meaning, similar to the wh-movement analysis for wh-fronting questions mentioned above.
Despite the research looking into the similarities and differences between wh-fronting and wh-in-situ questions, including their interpretations and restrictions, the question of whether one language can use both of these strategies is still under debate. Many languages that seem to allow both types of questions are argued to only allow wh-movement or wh-in-situ (LF movement/unselective binding), upon closer scrutiny. For example, Cheng (Reference Cheng1991) argues that Indonesian is a wh-in-situ only language, and the wh-fronting questions in Indonesia are cleft sentences not derived by wh-movement (cf. Cole & Hermon Reference Cole and Hermon1998). Faure & Palasis (Reference Faure and Palasis2021) argue that Colloquial French is a wh-in-situ language and movement of wh-items in wh-fronting questions is not triggered by a Question operator but rather an Exclusivity operator. From the other direction, Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) argue that in languages with wh-movement, wh-in-situ questions are not derived via LF movement or unselective binding, unlike Chinese and other ‘true’ wh-in-situ languages. For example, while North American English primarily uses wh-fronting, cases of wh-in-situ do appear, as is shown in (5b) (Ginzburg & Sag Reference Ginzburg and Sag2000).
Although Example (5) might be perceived as English allowing both strategies, Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) propose that the apparent wh-in-situ question in (5b) is a ‘declarative syntax question’ (DSQ), with the syntax of a declarative sentence as shown in (6). There is no C with the [+Q] feature in the structure, unlike the wh-fronting or the wh-in-situ questions in English, Mandarin and CSE. The question interpretation of DSQs results entirely from pragmatic mechanisms (see also Ginzburg & Sag Reference Ginzburg and Sag2000).
One definitive diagnostic that can tease apart DSQs and true wh-in-situ questions is that DSQs cannot be embedded under question-selecting predicates like wonder, ask and want to know to form an indirect question. These predicates require the embedded CP to have a [+Q] feature, and DSQs lack [+Q] on their C heads. The sentences in (7) show that the so-called wh-in-situ questions in English indeed cannot be embedded under wonder, indicating that they are DSQs, while wh-fronting questions have no such restrictions, confirming the presence of [+Q]. The sentence in (8) shows that wh-in-situ questions in Mandarin can be embedded under want to know, thus are not DSQs. According to Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015), languages that pattern with English include German, Dutch, Icelandic, American Sign Language, Brazilian and European Portuguese, while Turkish, Chinese and Japanese are true wh-in-situ languages.
Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) further propose the generalization in (9): In a language that allows wh-fronting questions, the seemingly wh-in-situ questions must be DSQs. In other words, it is impossible for one language to contain both wh-movement and true wh-in-situ questions (be it covert wh-movement or unselective binding).
If the generalization in (9) is on the right track, languages that seemingly allow both wh-fronting and wh-in-situ questions can be classified into two groups. Group 1 (true wh-movement) allows wh-movement only, and not covert wh-movement or unselective binding. The wh-in-situ questions in these languages are necessarily DSQs. Group 2 allows true wh-in-situ questions (derived via covert wh-movement or/and unselective binding). The wh-fronting questions in this group are necessarily derived via other operations (e.g. cleft, focus movement, etc.). Indonesian (following Cheng Reference Cheng1991) and Colloquial French (following Faure & Palasis Reference Faure and Palasis2021) would belong to this group, as their wh-fronting questions are argued to be not derived by wh-movement. On the other hand, to falsify the generalization, one would show that a language allows both wh-fronting questions derived by wh-movement and wh-in-situ questions that can be embedded under question-selecting predicates. This paper argues that CSE fits this profile.
CSE is an interesting test case since alternative analyses of both kinds of wh-questions in CSE have been proposed, potentially corroborating the generalization. Chang (Reference Chang2016) proposes that wh-in-situ questions in CSE are indeed DSQs, instead of true wh-in-situ questions (cf. Sato & Ngui Reference Sato and Ngui2017). Lan (Reference Lan2016), on the other hand, proposes that the wh-fronting questions in CSE do not involve wh-movement, but rather are cleft sentences, similar to Bahasa Indonesia (Cheng Reference Cheng1991). If either of these proposals is correct, CSE would conform to the generalization in (9).
This paper uses methods from experimental syntax and shows that predictions from the two alternative analyses for wh-questions in CSE are not borne out. This indicates that CSE indeed allows both wh-question strategies, challenging the DSQ/wh-in-situ generalization. Regarding methodology, setting the empirical record straight for a contact language like CSE can be tricky, as it is subject to a greater degree of individual variation. While previous studies in CSE have utilized methods from experimental syntax (Chang Reference Chang2016, Sato & Ngui Reference Sato and Ngui2017), experiments in the current paper feature the factorial design which controls for potential confounds and is widely used in experimental syntax.
The rest of the paper will proceed as follows. Section 2 provides a basic introduction of CSE. Section 3 argues for the availability of true wh-in-situ in CSE. Section 4 argues for the availability of true wh-fronting in CSE. Section 5 discusses theoretical implications.
2. Colloquial Singapore English
CSE, also known as Singlish, is a contact language with a dominant English lexifier and is strongly influenced by its various substrate languages, including Mandarin Chinese, Malay and local Sinitic languages. CSE allows both wh-fronting and wh-in-situ as is shown in (10). The options would present an apparent counter-example to Bobaljik & Wurmbrand’s generalization, if the wh-fronting question in (10a) involves wh-movement and the wh-in-situ question in (10b) is not a declarative syntax question. For more work on wh-questions in CSE, see Chow (Reference Chow1995), Bao (Reference Bao2001), Kim et al. (Reference Kim, Chang, Lau and Nagarajan2009), Yeo (Reference Yeo2010), Sato (Reference Sato2013) among others.
However, two analyses of CSE questions can, in principle, salvage the generalization. The first analysis is one where the wh-in-situ questions in CSE are indeed DSQs, proposed by Chang (Reference Chang2016). This would put CSE in the camp of varieties of English that are considered more ‘inner circle Englishes’. The second analysis is put forward by Lan (Reference Lan2016), where the wh-fronting questions in (10a) do not involve wh-movement but are analyzed as cleft sentences where the wh-element is base-generated at the beginning of the sentence. This puts CSE together with Mandarin and other varieties of Chinese, that is, languages that only allow wh-in-situ and not wh-fronting questions.
We will discuss three experiments in CSE showing that neither of the analyses is supported: the wh-in-situ questions in CSE do not show the same distribution as DSQs and the wh-fronting questions in CSE do not show properties of cleft sentences. Thus, both the DSQ analysis and the cleft analysis for CSE are challenged. The generalization in (9), in turn, is also challenged as CSE does allow both sc-movement and true wh-in-situ questions.
3. True wh-in-situ in CSE
This section discusses the DSQ analysis for wh-in-situ questions in CSE and argues that DSQ cannot be the only source for wh-in-situ questions in CSE. In other words, ‘true’ wh-in-situ questions do exist in CSE.
3.1. Declarative Syntax Question analysis for wh-in-situ questions in CSE
As mentioned above, non-echo wh-in-situ questions have been observed in English, German and other wh-fronting languages (Ginzburg & Sag Reference Ginzburg and Sag2000, Bobaljik & Wurmbrand Reference Bobaljik, Wurmbrand, Gallego and Ott2015), shown in (11).
Although on the surface, these questions are similar to true wh-in-situ questions in Chinese and Japanese, Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) observe that the questions in (11) cannot be embedded under question-selecting predicates like wonder or ask. In other words, wh-in-situ questions in (11) cannot function as indirect questions as shown in (12). The same restriction is found in German, Dutch, French, Icelandic, (Brazilian) Portuguese and American Sign Language.
Importantly, wh-in-situ questions in Chinese and other ‘classic’ wh-in-situ languages do not have this restriction, as shown in (13). wh-in-situ questions can be embedded under a question-selecting predicate.
Based on this contrast, Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) propose that the questions in (11) have a different structure from the ‘true’ wh-in-situ questions in (13). The question-selecting predicate selects a CP that is specified as [+Q]. The incompatibilities in (12) indicate that the wh-in-situ questions in (12) are not [+Q]. Instead, these embedded ‘questions’ have exactly the same syntax as a declarative sentence. The question meaning of such DSQs results from a pragmatic process.
Chang (Reference Chang2016) proposes that CSE is another language that only allows wh-fronting and does not allow true wh-in-situ. The seemingly wh-in-situ languages in CSE are declarative syntax questions. The proposal is built on the observation that wh-in-situ questions under question-selecting predicates are degraded compared to the wh-fronting versions, as is shown in (14). The contrasts in (14) are based on two surveys that Chang conducted with 18 and 10 speakers, respectively.
However, the robustness of the contrast is debatable. Sato & Ngui (Reference Sato and Ngui2017) argue that both wh-in-situ and wh-fronting questions are allowed under question-selecting predicates in CSE. They conducted a survey with the sentences in (15) with 13 CSE speakers, and the majority (11/13) of the speakers found all of them acceptable. In other words, no contrast was found in this survey. Based on this result, Sato & Ngui (Reference Sato and Ngui2017) argue that CSE falsifies the cross-linguistic generalization from Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015), contra Chang (Reference Chang2016).Footnote 1
Although the results are distinct, the surveys in Chang (Reference Chang2016) and Sato & Ngui (Reference Sato and Ngui2017) share a flaw in their design. The surveys assume that if wh-in-situ questions are dispreferred as opposed to wh-fronting questions under question-selecting predicates, the wh-in-situ questions in CSE are incompatible with question-selecting predicates. However, there is another way to interpret the contrast in (14). In a scenario where wh-in-situ questions are allowed under question-selecting predicate, but there is a general dispreference of wh-in-situ in CSE, we would still expect the contrast reported in (14). In other words, the contrast could be between wh-in-situ and wh-fronting in general, not related to question-selecting predicates. This confound casts doubts on conclusions in both Chang (Reference Chang2016) and Sato & Ngui (Reference Sato and Ngui2017) as they share the same design.
In the next section, we report an experiment with a 2*2 factorial design that is free of this confound. Our results show that contrary to Chang (Reference Chang2016) and compatible with Sato & Ngui (Reference Sato and Ngui2017), wh-in-situ questions under question-selecting predicates do not induce degradation in judgments, compared to wh-fronting questions under question-selecting predicates. Thus, the DSQ analysis cannot be the only source for wh-in-situ questions in CSE. It is important to note that we are not arguing that DSQs do not exist in CSE. We assume the English-type DSQs exist in most, if not all, languages. Rather, our findings show that not all wh in-situ questions in CSE are DSQs. Thus, the mechanism behind true wh-in-situ must exist in CSE.
3.2. Experiment 1: Embedded wh in-situ under question-selecting predicates
3.2.1. Design
Experiment 1 investigates whether wh-in-situ questions can be embedded under question-selecting predicates in CSE. It has two factors: wh-strategy (move vs. in-situ) and embedding (matrix vs. embedded). An example of each condition is shown in (16). The questions in the embedded conditions are embedded under the question-selecting predicate ‘want to know’. Sentences in the matrix conditions are matrix questions with an extra prepositional phrase modifier (for dinner in (16)) to keep the length of the sentences more similar to the other conditions.
In this design, the acceptability difference between move.matrix and in-situ.matrix (D1) is driven by the general preference between the two wh-strategies. The difference between move.embedded and in-situ.embedded (D2) also includes the effect of this general preference. If there is an additional penalty of embedding wh-in-situ under question-selecting predicates (i.e. if DSQ is the only source for wh in-situ in CSE as is claimed by Chang Reference Chang2016), D2 would also include this penalty. As a result, D2 should be notably larger than D1, and the statistical analysis should reveal a statistically significant interaction between the two factors. Moreover, the in-situ.embedded condition should have a low rating as it is predicted to be unacceptable like (12) under the DSQ analysis. Note that the 2*2 factorial design controls for the general preference between the two wh-strategies that the previous surveys are confounded with: both D1 and D2 are affected by this preference, the comparison between D1 and D2 would cancel out its effect. This factorial design is widely used in experimental syntax literature to control for such confounds, especially in locality constraints like syntactic islands (see Sprouse et al. Reference Sprouse, Caponigro, Greco and Cecchetto2016 among others).
3.2.2. Materials, procedures and participants
Eight lexical combinations were created for each condition, resulting in 32 test items in total. The test items are distributed in a Latin Square design. Each participant saw two items per condition. Each list also includes eight test sentences from Experiment 2 and 12 additional filler items. Experiments 1 and 2 were conducted in the single session with their test items intermixed with the filler items, thus their procedure and participants are identical. The experiment was conducted on PCIbex (Zehr & Schwarz Reference Zehr and Schwarz2018), where each participant was asked to rate how natural the sentences sound as CSE on a 7-point Likert scale, 1 being very unnatural and 7 being very natural. The instructions emphasized that the experiment is meant for CSE and not standard inner circle Englishes. The test sentences do not have tense or agreement marking as shown in (16), which is allowed in CSE but not standard inner circle Englishes. Thirty-six participants finished the experiment, all of whom were aged from 21 to 30, grew up in Singapore and use CSE on a daily basis. Each participant was compensated with 5 SGD for participation.
3.2.3. Results and analyses
The 7-point ratings were transformed to z-scores in order to control for individual bias in using the scale. Table 1 summarizes the condition means and standard deviations in raw ratings and z-scores. Each condition mean is based on 72 judgments. The condition means in raw ratings and the standard errors are plotted in Figure 1. As we can see, all four conditions are rated around 6 out of the 7-point scale, and all are above 0.6 in z-scores. Figure 2 shows the distribution of condition means by subjects. Judgments of all four conditions show a normal distribution, indicating that the judgments are largely homogeneous with moderate speaker variation.
Table 1. Results from Experiment 1, n = 36.
Figure 1. Condition means of raw judgment of Experiment 1, N = 36.
Figure 2. Distribution of condition means by subjects for Experiment 1. Condition means for each participant were calculated based on their two judgments of the conditions.
We constructed a cumulative link mixed model on raw judgments using embedding and whStrategy as fixed factors and items and participants included as random factors. Treatment coding was applied to both factors (matrix: 1, embedded: –1; move: 1, situ: –1). It revealed no main effects of embedding (p = .11) but a main effect of whStrategy (p = .02). There is also no interaction between the two conditions (p = .17). The results are summarized in Table 2.Footnote 2
Table 2. Results from the cumulative link mixed model, formula: judgment ~ embedding * whStrategy + (1 + embedding * whStrategy | subject) + (1 + embedding * whStrategy | lexical).
Note that the acceptability of embedded wh-in-situ questions in this experiment cannot be due to participants parsing them as echo questions. As echo questions are questions about a declarative sentence in the previous utterance, to parse B’s utterance in (17) as an echo question, A’s utterance needs to be a declarative sentence. This cannot be the case given the question-selecting predicate, the declarative sentence in A’s utterance in (17) is itself ungrammatical. In other words, if the only sources for wh-in-situ questions in CSE are DSQs and echo questions, the acceptability of the in-situ.embedded condition is unexpected.Footnote 3
Given the high ratings of the in-situ.embedded condition and the lack of interaction between the two factors, results from Experiment 1 show wh-in-situ questions can be embedded under question-selecting predicates in CSE, contrary to the prediction of the DSQ analysis. Thus, DSQ cannot be the only source for wh-in-situ in CSE. In other words, true wh-in-situ questions do exist in CSE. Note that our conclusion is that the DSQ analysis cannot be the only source for wh-in-situ questions in CSE, not that DSQs do not exist in CSE.
3.3. English experiment on DSQs for comparison
Although no statistically significant factor was found on embedded wh-in-situ questions in CSE, the in-situ.embedded condition is rated lower than the other three conditions in Experiment 1. A reviewer asks if it is possible that the lack of interaction in Experiment 1 is due to the low number of participants (N = 36). To further verify the status of wh-in-situ questions in CSE, we conducted a similar experiment in North American English (NAE) with 35 participants for comparison. As laid out above, the DSQ analysis of wh-in-situ questions in NAE predicts that embedding such questions under question-selecting predicates would induce a clear penalty, as they are ungrammatical. If wh-in-situ questions in NAE are DSQs and wh-in-situ questions in CSE are true wh-in-situ questions, we would expect a statistically significant interaction between the two factors and a low rating of the embedded wh-in-situ condition in NAE with a similar number of participants, assuming similar levels of noise in the two languages.
This experiment uses the same design as Experiment 1. Two factors were manipulated: wh-strategy (move vs. in-situ) and embedding (matrix vs. embedded). Do support and tense markings were added to sentences in the English conditions. Some of the names were replaced with more frequent names in North America. Examples of each condition are shown in (18). The experiment includes two items for each condition in (18) (eight test items), eight test items for Experiment 2 in NAE and 12 fillers. Thirty-five self-reported English monolingual speakers participated in the experiment, recruited via Prolific. Each participant was paid 1.5 GBP for their participation.
Mean ratings of the four conditions in raw judgments and z-scores and their standard deviations are summarized in Table 3. Figure 3 shows the conditions means and standard errors in raw judgments, and Figure 4 shows the distribution of condition means by subjects. The move.matrix and move.embedded conditions were rated near ceiling. The in-situ.matrix condition was rated a slightly lower mean of 5.01 and showed a considerable amount of speaker variation. This is expected given that wh-in-situ is not a common strategy to form a question in NAE. Wh-in-situ questions are accepted as echo questions or DSQs. It is possible that some participants treated the in-situ.matrix condition as echo questions or DSQs and gave it a higher rating, while others treated it as (ungrammatical) common information-seeking questions and gave it a lower rating. Crucially, the in-situ.embedded condition is rated close to the bottom of the scale (mean = 2.30). This is expected since sentences in this condition cannot be treated as echo questions, or, according to Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015), DSQs, given the question-selecting predicates.
Table 3. Results from the NAE DSQ experiment, n = 35.
Figure 3. Condition means of raw judgments, N = 35.
Figure 4. Distribution of conditions means by subject for NAE DSQ experiment, N = 35.
We constructed a cumulative link mixed model to fit the raw judgments. Treatment coding was applied to both factors (matrix: 1, embedded: –1; move: 1, situ: –1). There was a main effect of whStrategy: conditions with wh-movement are rated higher than wh-in-situ. There was also a main effect of Embedding. Crucially, there was a significant interaction between the two factors: embedding wh-in-situ questions under question-selecting predicates induced an extra penalty. To our knowledge, this is the first experimental finding regarding the status of matrix and embedded wh-in-situ questions in NAE.
Table 4. Results from the cumulative link mixed model, formula: judgment ~ embedding * whStrategy + (1 + embedding + whStrategy | subject) + (1 + embedding * whStrategy | lexical).
Results from Experiment 1 and the NAE experiment with the same design and a similar number of participants show that wh-in-situ questions are treated differently in two languages. Although the mean ratings of in-situ.matrix conditions in both languages are above 5, the distribution of the ratings in CSE was homogeneous, while there was a significant amount of speaker variation in NAE. This indicates that wh-in-situ questions in CSE are common information-seeking questions that require no special contexts, while the wh-in-situ questions in NAE are likely DSQs and/or echo questions that do require contexts to be acceptable. Crucially, the in-situ.embedded condition in NAE is unacceptable with a mean rating of 2.3, which is compatible with the DSQ analysis. The same condition in CSE has a mean rating of 5.96, compatible with the true wh-in-situ analysis. We thus conclude that DSQs cannot be the only source for wh-in-situ questions, and true wh-in-situ questions exist in CSE.
4. wh-movement in CSE
4.1. The cleft analysis for wh-fronting questions in CSE
Having established that CSE allows true wh-in-situ questions, this section argues for the existence of true wh-fronting questions in CSE which are derived by wh-movement as in English. As mentioned in Section 1, wh-fronting questions are typically analyzed to result from the wh-element undergoing wh-movement to the SpecCP headed by a C with [+Q]. A sample derivation is shown in (19), where the wh-element moves from its base-generated position (object of the verb) to the SpecCP position.
However, the wh-cleft analysis has been proposed for Bahasa Indonesia (Cheng Reference Cheng1991), Mandarin Chinese (Cheung Reference Cheung2014) and CSE (Lan Reference Lan2016) as an alternative analysis for wh-fronting questions. Under this analysis, the seemingly ‘fronted’ wh-element is base-generated at its surface position. In other words, no wh-movement is involved in deriving wh-fronting questions. We will go over the wh-cleft analysis in this section, and the next section reports two experiments that show wh-cleft analysis for wh-fronting questions in CSE is untenable.
Focusing on CSE, Lan (Reference Lan2016) proposes that the wh-fronting questions in CSE are derived based on it-cleft sentences where the wh-word is base-generated at SpecFocusP as is shown in Example (20). The wh-word is coindexed with an empty operator (op) at the object position. Several elements, including it, is and that are elided, resulting in what seems to be a question with wh-movement. In other words, the wh-fronting questions in CSE, although similar to those in inner circle Englishes on the surface, are derived in a different way, with no wh-movement of the wh-element.
Lan (Reference Lan2016) provides two arguments for this analysis. The first argument comes from the absence of superiority effects in wh-fronting questions in CSE. Superiority effects in North American English are shown in (21). When there are multiple wh-elements in a sentence, only the hierarchically higher element can be fronted, for example, who in (21a). Moving the lower wh-element across the higher one induces unacceptability, for example, what in (21b). This is a general constraint on wh-movement based on the hierarchical relation between the two wh-elements (see Chomsky Reference Chomsky, Anderson and Kiparsky1973, Pesetsky Reference Pesetsky2000, Richards Reference Richards2001 among others). The presence of such an effect is considered to be evidence for wh-movement of the fronted elements in (21).
Lan (Reference Lan2016) provides the examples in (22) to argue that wh-fronting questions in CSE do not show superiority effects. Example (22a) is a baseline sentence where the highest wh-element who is fronted, predicted to be acceptable. In (22a), where, which is base-generated below the subject who, is fronted to the sentence-initial position. If the wh-fronting question in (22b) involves wh-movement of where, it is predicted to show superiority effects like (21b). The acceptability of (22b) thus indicates that where did not undergo movement across other higher wh-elements. The cleft analysis, on the other hand, can account for the lack of superiority effects as is shown in (23): the ‘fronted’ where is base-generated at its surface position, no superiority effects are induced as no movement across a higher wh-element occurred.
The second argument for the cleft analysis comes from the optional is at the sentence-initial position of wh-fronting questions. Lan (Reference Lan2016) reports the contrast in (24). An optional is can appear at the beginning of a wh-fronting question (Example 24a) but not a wh-in-situ question (Example 24b).
The cleft analysis can derive this pattern straightforwardly since wh-fronting questions are derived from it-cleft sentences and the ellipsis of it, is and that. The sentence in (24a) can be derived if only it and that are deleted and is remains as shown in (25). Wh-in-situ questions are not derived from cleft sentences in the first place, so the presence of is is not grammatical.
If the cleft analysis is the only source for wh-fronting questions, CSE would be a wh-in-situ language like Mandarin Chinese, with a non-wh-movement derivation for the wh-fronting questions, thus conforming to the generalization from Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015) that only one wh-strategy can be available in any given language. Given these two arguments for the cleft analysis for wh-fronting questions, Experiments 2 and 3 test the lack of superiority effects and the availability of sentence-initial is, respectively. Note that we test the strong claim that the cleft analysis derives all cases of wh-fronting questions in CSE, leaving no space for wh-movement; rather than the weak position that the cleft analysis and the wh-movement analysis are both possible in CSE. We do not aim to exclude the possibility of some wh-fronting questions being derived from cleft sentences.
4.2. Experiment 2: Superiority
Experiment 2 tests whether wh-fronting questions in CSE show superiority effects. According to the cleft analysis specified above, CSE should not show superiority effects. On the other hand, if wh-fronting questions do involve wh-movement, superiority effects are predicted.
4.2.1. Design
Experiment 2 includes two factors: (1) wh count, whether the sentence includes one or two wh-elements (single vs. multiple); (2) word order, whether the subject precedes the object or vice versa (subj-obj vs. obj-subj). A sample of the four conditions is shown in (26). In single conditions, either the subject or the object is a wh-element, while in multiple conditions, both of them are wh-elements. In subj-obj conditions, the subject precedes the object, and in obj-subj, the object is moved across the subject to produce the object-subject order. Note that in order to keep the sentences simple and closer to the conventional cases of superiority effect, we used subject and object wh-questions and not adjunct questions (e.g. where).
Another choice we made is to embed the questions under a matrix clause (you think … in (26)), rather than testing matrix questions shown in (27). This decision was made to make sure that the multiple.obj-subj is not dispreferred due to the two adjacent wh-elements at the sentence-initial position (what who in (27d)). Note that this partially results from the fact that CSE speakers prefer to drop the auxiliaries in wh-questions, (27d) would have been What did who order yesterday? in inner circle Englishes.
The acceptability difference between single.subj-obj and single.obj-subj (D1) would result from the difference between a subject-fronted question and an object-fronted wh-question. The difference between multiple.subj-obj and multiple.obj-subj (D2), both of which are multiple questions, would also include the difference between a subject-fronted question and an object-fronted wh-question. Note that the effect of being a multiple-wh question would be canceled out, given that both multiple.subj-obj and multiple.obj-subj conditions involve multiple-wh questions. If wh-fronting is driven by wh-movement, the multiple.obj-subj condition involves what moving across the higher wh-element who, which would induce the superiority condition. The multiple.subj-obj condition, although also a multiple wh-question, should not induce superiority effects since the fronted wh-element is the subject which is base-generated at a higher position than the object. As a result, according to the wh-movement analysis, D2 would additionally include the penalty from superiority effects.
If wh-fronting questions in CSE do involve wh-movement, it would predict (26d) to be ungrammatical, that is, showing superiority effects. Statistical tests should reveal a significant interaction between the two factors: D2 should be larger than D1, as D2 includes superiority effects. If wh-fronting questions in CSE do not involve wh-movement and the fronted wh-elements are base-generated, (26d) should be acceptable. D1 and D2 should be of the same size, as they are both driven by the same factors specified above. No interaction between the two manipulated factors is predicted.
4.2.2. Materials, procedures and participants
Eight lexical combinations were created for each condition, resulting in 32 test items in total. The test items are distributed in a Latin Square design. Each participant saw two items per condition. Each list also includes eight test sentences from Experiment 1 and 12 additional fillers. Experiments 1 and 2 were conducted together, thus the procedure and participants are identical to Experiment 1. Thirty-six CSE speakers finished the experiment. All test items in Experiment 2 end with the sentence-final particle ah indicating the question’s meaning. The sentence-final particle ah was included in order to further make sure that participants are judging the CSE sentences and not inner circle Englishes, as sentence-final particles like ah are only allowed in the former.
4.2.3. Results and analyses
Table 5 summarizes the condition means of Experiment 2 in raw judgments on the 7-point scale and transformed z-scores, as well as their standard deviations. Three conditions that do not involve superiority effects (single.subj-obj, single.obj-subj, multi.subj-obj) are rated above the midpoint of the scale for raw score (4 out of 7) and for z-score (0). The multi.obj-subj condition is rated below 4 and with a negative z-score.
Table 5. Results from Experiment 2, n = 36.
The condition means in raw ratings and their standard errors are plotted in Figure 5. Figure 6 shows the distribution of condition means by subjects. The difference in judgment between single.subj-obj and single.obj-subj (D1) is 0.19 in raw ratings and 0.1 in z-score. The difference between multi.obj-subj and multi.obj-subj conditions (D2) is 1.79 in raw ratings and 0.787 in z-score. D2 is clearly larger than D1. In other words, the multi.obj-subj condition involves an extra penalty that does not affect the other three conditions.
Figure 5. Condition means of raw judgments of Experiment 2, n = 36.
Figure 6. Distribution of raw judgments of Experiment 2, n = 36.
We constructed a cumulative link mixed model on raw judgments using wh count and word order as fixed factors with items and participants included as random factors. Treatment coding was applied to both fixed factors (single:1, multiple:-1; subj-obj:1, obj-subj:-1). The model revealed a significant main effect of wh count (p
$ < $
.0001): multiple wh-questions are rated lower than single wh-questions. There was also a significant main effect of word order (p
$ < $
.001). Crucially, there is a significant interaction between the two factors (p
$ < $
.001). The results from the statistical tests are summarized in Table 6. The statistically significant interaction indicates that multi.obj-subj is dispreferred to multi.subj-obj more than single.obj-subj is dispreferred to single.subj-obj.
Table 6. Experiment 2: results from the cumulative link mixed model, judgment ~ order * whCount + (1 + order * whCount | subject) + (1 + order * whCount | lexical).
The low rating of the multi.obj-subj condition and the significant interaction between the two factors are compatible with superiority effects in wh-fronting questions in CSE. As laid out in the previous section, this is expected under the wh-movement analysis for wh-fronting questions, as superiority effects result from a constraint on wh-movement. The results are not compatible with the cleft analysis, where the fronted wh-element is base-generated in its surface position.
4.3. Experiment 3: is in wh-fronting questions
Experiment 3 tests the other prediction of the cleft analysis: the availability of optional is in wh-fronting questions. According to the cleft analysis, wh-fronting questions in CSE are derived from it-cleft sentences with deletion of it, is and that. Lan (Reference Lan2016) claims that CSE also allows leaving is undeleted. This derivation (see 24–25) predicts that is can appear in the sentence-initial position in wh-fronting questions but not in wh in-situ questions.
4.3.1. Design
Experiment 3 includes two factors: whStrategy (fronting vs. in-situ) and whether is is present (labeled as isPresence) (present vs absent). Sample items of the four conditions are shown in (28).
The cleft analysis predicts that the wh-fronting question with is present (28c) is acceptable while the presence of is is incompatible with wh-in-situ questions as in (28d). The absence conditions were included to control for baseline preference between wh-fronting and wh in-situ questions in CSE (see discussion in Experiment 1). The difference between (28a) and (28b) (D1) results from a general preference between these two strategies. If the cleft analysis is on the right track, the difference between (28c) and (28d) (D2) includes this general preference and the extra penalty of having an is at the beginning of a wh-in-situ question. Example (28c) should not be affected by this penalty, as is is claimed to be compatible with wh-fronting questions. Note that both (28c) and (28d) contain an is at the sentence-initial position, thus the general effect of including is is canceled out. With this design, the cleft analysis predicts that D2 is larger than D1 since (28a-28c) are grammatical while (28d) is not. Statistically, this would be reflected by a statistically significant interaction of the two factors.
4.3.2. Materials, procedures, participants
Eight lexically matched sets were constructed for each condition, resulting in 32 test items in total. Each participant saw two items per condition (eight test items in total). Each test item is from a different lexically matched set. Each participant also saw 16 filler items.
The participants were asked to judge how natural the sentences were on a 7-point scale, 1 being completely unnatural and 7 being completely natural. The participants were instructed to rate the sentences based on their intuition of CSE. All test items end with a CSE sentence-final particle ah. The experiment was conducted on PCIbex (Zehr & Schwarz Reference Zehr and Schwarz2018). Thirty-two self-identified CSE native speakers participated in Experiment 3, with ages ranging from 21 to 35. They were recruited from personal contacts via social media messaging services. Participants were not compensated for participating in Experiment 3.
4.3.3. Results and analyses
The 7-point scale judgments were z-score transformed. The condition means in raw judgments and z-scores are included in Table 7 along with their standard deviations. The condition means of raw judgments and their standard errors are plotted in Figure 7. Figure 8 shows the distribution of condition means by subjects. The results show that the conditions where is is present are rated below the midpoint of the scale, while the absent conditions are rated above it. Both wh-in-situ and wh-fronting questions were judged toward the bottom of the scale in the presence of is. This is unexpected if the sentence-initial is is compatible with wh-fronting questions as predicted by the cleft analysis.
Table 7. Results of Experiment 3, n = 32.
Figure 7. Condition means of raw judgments of Experiment 3, n = 32.
Figure 8. distribution of raw judgments of Experiment 3, n = 32.
We constructed a cumulative link mixed model on raw judgments with isPresence and whStrategy as fixed factors and with items and participants as random factors. Treatment coding was applied to both fixed factors (absent:1, present:-1; move:1, situ:-1). We started with a model that included all random slopes, but due to a failure to converge, the random effects structure was simplified until the largest converging model was achieved (see Table 8 for the final model). The results from the statistical tests are summarized in Table 8. The model revealed a significant main effect of isPresence (p
$ < $
0.0001), conditions with is are rated worse than the ones without. There is a marginal main effect of whStrategy (p = 0.07). There is no interaction between the two factors (p = 0.30).
Table 8. Formula: Experiment 3 results, formula: judgment ~ isPresence * whStrategy + (1 + isPresence | subject) + (1 + isPresence * whStrategy | lexical).
Results from Experiment 3 show that for most of the participants, the presence of is makes both wh-fronting and wh-in-situ questions unacceptable. The predicted interaction of the two factors by the cleft analysis is not observed. The size of the penalty of including is at the beginning of questions is not smaller for wh-fronting. In fact, the effect of is is numerically larger for wh-fronting questions (3.58 in raw judgment) than for wh-in-situ questions (2.72 in raw judgment). This is the opposite of what the cleft analysis predicts and could result from the fact that wh-fronting questions with no is are rated higher than wh-in-situ questions. On the other hand, the wh-movement analysis would account for the effect of is: neither wh-in-situ nor wh-fronting questions are derived from it-clefts, thus is is never generated at the question initial position.
We note that Figure 8 shows a bimodal distribution for the fronting.present condition where eight out of the 32 participants rated the condition between 4 and 6 while the rest of the participants rated it at the bottom of the scale. This pattern indicates that for a small portion of participants, the sentence-initial is is acceptable in wh-fronting questions. Given the consistently low ratings for the in-situ.present condition, these eight participants did show a preference for fronting.present sentences, as predicted by the cleft analysis for wh-fronting questions. Thus, we leave open the possibility that a portion of CSE speakers can assign a cleft structure to wh-fronting questions. However, this possibility is compatible with our claim that wh-movement exists in CSE, as the existence of wh-movement does not exclude the possibility of some wh-fronting questions being derived from cleft sentences. Crucially, the majority of our participants rated the fronting.present condition at the bottom of the scale, which indicates these participants did not assign the cleft analysis to the wh-fronting questions. In other words, the cleft analysis cannot be the only source for wh-fronting questions in CSE. Thus, we maintain our conclusion that for CSE speakers, wh-fronting questions can involve wh-movement.
Combining Experiments 2 and 3, we can see that for most participants, neither of the arguments for the cleft analysis is verified in CSE. On the other hand, results from both experiments are consistent with the wh-movement analysis for wh-fronting questions, especially the presence of superiority effects in Experiment 2. Moreover, the distribution of the judgments from these experiments indicates that wh-movement is the primary source for wh-fronting questions in CSE.
Since the cleft analysis of wh-fronting questions has been proposed for Malay and Bahasa Indonesia (Cheng Reference Cheng1991), arguing against this analysis as the primary source for CSE wh-fronting questions corroborates with the view that vernacular varieties of Malay are not primary substrates of CSE (see Sato Reference Sato2013). Instead, wh-fronting questions and wh-in-situ questions rise under the grammatical pressure from inner circle Englishes and Chinese.
A reviewer observed that sentences in the absent conditions in Experiment 3 are different from the matrix conditions in Experiment 1 in that the former used where and the latter used what as the wh-elements. Examples of the relevant conditions and their mean ratings are repeated below in (29). The reviewer pointed out that there seems to be a penalty in embedded in-situ questions with where: (29b) is rated lower than (29a), but no difference is observed in (29c–29d). Another reviewer reported, based on their own judgments in CSE, that ‘Yesterday Charles eat laksa where ah?’ or ‘Charles eat laksa where ah?’ sound better than (29b) and suggests semantic/pragmatic effects at play here.
We note the observations but want to highlight that the ‘where’ was chosen to stay close to the original claim by Lan (Reference Lan2016), who used questions with ‘where’ (see (24)). While a potential penalty on in-situ questions in (29b) is intriguing in its own right, we do not think it would compromise our conclusion in Experiment 3. The cleft analysis predicts that wh-fronting questions with a sentence-initial is are acceptable, and our results show that fronting.present was rated at the bottom of the scale and clearly lower than the fronting.absent condition.
5. General discussion
Experiment 1 shows that both wh-in-situ and wh-fronting questions can be embedded under question-selecting predicates in CSE. This result indicates that declarative syntax questions cannot be the sole source of wh-in-situ questions in CSE. ‘True’ wh-in-situ questions, derived either by unselective binding or covert movement, are available in CSE as they are in languages like Mandarin Chinese.Footnote 4 Experiment 2 verifies the presence of superiority effects in wh-fronting questions in CSE, and Experiment 3 shows that the presence of is at the beginning of questions is generally unacceptable in CSE, challenging the cleft analysis for the wh-fronting questions. Both Experiments 2 and 3 point to the presence of wh-movement in CSE.
Taking these results together, both ‘true’ wh-in-situ and wh-movement questions exist in CSE, as the alternative analyses proposed in the literature are not supported (Lan Reference Lan2016, Chang Reference Chang2016). This makes CSE a language that is at odds with the generalization that each language can only allow one wh-strategy proposed in Bobaljik & Wurmbrand (Reference Bobaljik, Wurmbrand, Gallego and Ott2015). As mentioned in Section 3, Sato & Ngui (Reference Sato and Ngui2017) argued that CSE allows both ‘true’ wh-fronting and wh-in-situ questions as well, based on their survey on DSQ. However, Sato & Ngui (Reference Sato and Ngui2017) did not address the cleft analysis (Lan Reference Lan2016), and their methodology suffers from the same confound as Chang (Reference Chang2016). The current paper uses more controlled experimental designs to verify the empirical claims and predictions generated by the theories under debate. As mentioned in Section 1, the existence of languages like CSE presents a counterexample to the DSQ/wh-in-situ generalization in Example (9). To accommodate languages that allow both wh-strategies is rather straightforward. Sato & Ngui (Reference Sato and Ngui2017) propose that two variants of the C head are present in such languages, one with a strong Q feature (C
$ {}_{\left[+ Qs\right]} $
in (30a)) which would require wh-movement of the wh-element to the SpecCP position, and one with a weak Q feature (C
$ {}_{\left[+ Qw\right]} $
in (30b)), which does not trigger wh-movement, and binds the wh-element in-situ instead (see also Cole & Hermon Reference Cole and Hermon1998 for a system proposed for Malay). There is no a priori reason to assume that the two versions of C heads cannot exist in the same language. Intuitively, having both variants in one language does not pose difficulties for acquisition either, as the positive evidence of wh-fronting and wh-in-situ questions are readily available in the input.
It is important to note that although this paper argues that wh-optionality between wh-movement and true wh-in-situ exists, it does not necessarily mean that the two operations are in free variation. It is an open question whether wh-movement and wh-in-situ questions have different pragmatic or semantic effects and are preferred in different contexts, even though both strategies are derived with the help of a C
$ {}_{\left[+Q\right]} $
head. Probing such intricate differences among wh-strategies is beyond the scope of this paper.Footnote
5
In terms of methodology, this paper presents an example of applying experimental syntax methods to contact and colloquial languages. Contact and colloquial languages like CSE are often reported to involve a considerable amount of speaker variation, which is why previous studies established their empirical claims with surveys with dozens of speakers, instead of relying on the authors’ own judgments. However, increasing the number of speakers itself does not guarantee more reliable data if the design is confounded. Using factorial designs for embedded questions, superiority effects and the presence of is in our experiments limits the space for confounding factors, thus lending more confidence in the empirical claims than the previous literature.
Acknowledgements
We thank the editor and the reviewers for their valuable comments and suggestions. Thanks also go to Michael Yoshitaka Erlewine and Nick Huang for their feedback. Beth Chan contributed substantially to the experiments.
Data availability statement
The stimuli, data for the test and filler items, and analyses for the four experiments reported in this paper can be found here: https://osf.io/fknbx/?view_only=c90ccb94b7d9418abcd727305019c72d
Open access
