Linking French wh-in-situ and context: a choice function analysis

4.1 The extra-strong presupposition

The so-called ‘extra-strong presupposition’ is usually attributed to Chang (1997), although she cites Coveney (1989a). It is illustrated by (26) and (27), which display a wh-fronted and a wh-in-situ question, each accompanied by an answer.

The wh-fronted question in (26) is a neutral question. Like other wh-questions, it involves an existential presupposition or implicature: the speaker expects there to be an answer to the question. The question in (26) can potentially receive a negative answer like rien ‘nothing’ (although this answer would not be expected). In contrast, the wh-in-situ question in (27) has been claimed to be felicitous only if the speaker already assumes that Marie bought something, since it is ‘strongly presupposed’ that there exists a value to fill the wh-phrase. The speaker merely requests more detail about the purchase, as in: what it is that Marie bought. Therefore, a negative reply like rien ‘nothing’ is quite odd.

We suggest that the questions that were claimed to exhibit an extra-strong-presupposition involve choice function wh-in-situ. To license (27) as choice function wh-in-situ, that is, for the value of the choice function to be recoverable, the referent for the wh-phrase must be identifiable in the preceding context: the entity that Marie bought must already be salient for the speaker. The speaker merely requests further information about this entity. The reply that Marie bought nothing would then not only be an unexpected answer but also an incongruent one.

Another example that has been claimed to exhibit an extra-strong presupposition is given in (28). This example illustrates the relevance of the speaker’s associations based on world knowledge for this alleged stronger presupposition.

Speaker A mentions that Pierre’s birthday is coming up. Based on speaker B’s knowledge of the world, this leads him/her to assume that there is a particular present that speaker A is planning to buy for Pierre. The question of what this present is, is merely ‘seeking details on an already established (or presupposed) situation’ (Chang 1997: 46). Under the current proposal, the context in (28) (in combination with world knowledge) can make the referent for the in-situ wh-phrase, the present that speaker A is planning to buy for Pierre, salient for speaker B, which makes the value of the choice function recoverable. To be precise, the relevant CSM is ‘semi-salient’ in this example (see also Section 4.3): only one possible interpretation of the context makes a specific referent for the wh-phrase salient, while an interpretation that does not license the choice function is also possible. On the interpretation of the context that makes salient a specific referent, the salient present is the unique established referent for the wh-phrase. Speaker B in (28) does not know to what type of present this referent corresponds and thus ignores which imaginable realisation of the single proposition is a true proposition in the actual world. By uttering (28), speaker B requests information about the identity of this previously established referent.

It was previously not clear what the conceptual status was of the extra-strong presupposition or where it might come from. Starke (2001) and Baunaz (2005, 2011, 2016) attribute the alleged presupposition to a particular feature that is part of the internal structure of a strongly presupposed in-situ wh-phrase, but not of a wh-fronted one. In Baunaz’s (2011, 2016) terminology, an in-situ wh-phrase with the feature ‘specific’ or ‘partitive’ carries a stronger presupposition than a ‘non-presuppositional’ wh-phrase. Likewise, Boeckx attributes the stronger presupposition to a definite presuppositional D feature on an in-situ wh-phrase (Boeckx 1999; Boeckx et al. 2001). These proposals place the alleged presuppositional nature of wh-in-situ, which at first sight seems like a semantico-pragmatic phenomenon, in the morphology (nano-syntax) of the wh-phrase. Our proposal is similar in the sense that it attributes a specific or definite-like meaning to the wh-phrase. However, our analysis relates this meaning to the lack of wh-movement and attributes it to a felicity condition. This ties the extra-strong presupposition, the fact that it is associated with in-situ wh-questions in particular, and the lack of intervention effects in such questions all together.

Mathieu (2002) describes the stronger presupposition in terms of background in contrast to focus. Chang (1997), on the other hand, mentions that ‘all information other than the questioned element is taken for granted’ and that ‘the information expressed by everything, except the wh-word is already a salient part of the previous discourse’ (p. 44). This intuition is made specific by the current proposal: the referent for the wh-phrase must be established in the context to make the choice function recoverable, predicting the ‘extra-strong presupposition’.

Recall from Section 3 that we assume a felicity condition on the use of a choice function (cf. Ionin 2006), according to which the choice function must be recoverable for the interlocutor. Ionin argues for the specific indefinite this (as in There is this book about questions you really should read) that it is subject to a felicity condition, in contrast to a definite article, which introduces a presupposition. Contrary to presuppositions, felicity conditions do not require maximisation. If one can use a definite article rather than an indefinite one, one will do so due to Maximise Presupposition (Heim 2008). However, specific indefinite this N can be replaced by an N without provoking a problem. Something similar can be observed in wh-in-situ questions. It does not seem to be the case that a context that permits the use of a choice function wh-in-situ question is incompatible with ordinary wh-questions. That is expected if it is a felicity condition that provokes the context dependency. In that sense, the phenomenon referred to as an ‘extra strong presupposition’ is a combination of the same existential presupposition that is commonly assumed for any wh-question and a felicity condition on the interpretation of the wh-phrase (via the contextually bound choice function).

4.2 Choice functions and apparent interveners

We argued in Section 3 that a choice function wh-in-situ question is unacceptable if the referent for the wh-phrase is not retrievable from the context, and that these sentences are not sensitive to intervention effects. On the basis of this, one might expect classical interveners to not induce infelicity; as long as the referent picked out by the choice function is recoverable by the context, the sentence should be felicitous. Nevertheless, the matter is more complicated. We show below that the class of interveners is not homogeneous: some interveners seem to make in-situ wh-questions unacceptable while others appear to be more permissive. In this section, we argue that differences within the class of interveners are due to differences in how difficult it is for speakers to think of a context that satisfies the recoverability condition associated with the choice function. Given this explanation, the unacceptability provoked by the presence of an intervener in choice function wh-in-situ is thus not an intervention effect (see (3) above), and the differences between (apparent) interveners follow from our analysis.

Certain expressions, such as focus adverbs (seulement ‘only’) and negative expressions ((ne) jamais ‘never’ and (ne) personne ‘no one’), are commonly assumed to result in unacceptability (e.g. Baunaz 2011; Engdahl 2006; Hamlaoui 2010; Mathieu 1999).

Other expressions are more controversial in that previous literature does not consistently treat them as interveners. This is particularly true of (ne) pas ‘not’ and several universal quantifiers (toujours ‘always’, tous les N ‘all the N’, tout le monde ‘everybody’ and floating tous ‘all’). The examples in (30) and (31) display contradictory judgements of sentences with the same intervener.

The availability of a contextually bound choice function sheds light on these controversies. First of all, if a sentence is presented to an informant in isolation (i.e. without context), the informant is free to construe their own context, which may or may not license the use of a contextually bound choice function. Crucially, for some interveners, it is easier to envisage a context that makes the choice function recoverable than for others. An example of an intervener where this is relatively easy is plusieurs N ‘several N’. The sentence in (32) with this intervener was judged acceptable (Adli 2006).^[13]

To license a choice function in (32), the context must make salient that there is a specific place where several oak trees have been felled. The speaker then inquires where this place is. The noun phrase plusieurs chênes ‘several oaks’ can be either interpreted collectively or distributively. If it is interpreted collectively, the oaks will be all felled in a single place. In this case, the interlocutor may quite easily imagine a context in which there is a salient, contextually given place where these trees have been felled.

In contrast to plusieurs N ‘several N’, it is more difficult to construe a context that establishes a referent for a question containing (ne) que ‘only’, as illustrated by (33):

We suggest that (33) is judged to be unacceptable when presented to informants in isolation because it is difficult to envisage the necessary context, as indicated by the #.^[14] For the choice function to be licensed in (33), the context must make salient that there exist particular circumstances under which Jean only spoke to Suzanne. It is not straightforward to envisage a context that makes this salient, as the restriction on the context is quite specific. Nonetheless, a question with the focus expression (ne) que ‘only’ can be made felicitous given an appropriate context; native speakers we consulted found the example in (34) acceptable.^[15] (The intended reading of (34) is that at the events at the end of the year, they did not speak to other people than Suzanne.)

Note that the context in (34) is much less common in the sense that it describes a very particular situation (three people only talking to a certain person and not to others in different circumstances) and a communicative setting that is not completely straightforward either. This illustrates the contrast with (32), for which both the situation (a collection of trees being felled) and the communicative context (a discourse referent corresponding to the place where it happened is salient in the context) are much more straightforward. In sum, the differences between apparent interveners can largely be explained by how easily one can construe a context that licenses the use of a choice function for a sentence containing that apparent intervener.

4.3 Choice functions and differences among speakers

There is also variation in judgements of sentences containing the same ‘intervener’, which seems to be due to differences between speakers.^[16] An example in which the same sentence receives contradictory judgements is given in (35) (see also (30) and (31) above).

The choice function mechanism can also shed light on such variation. As was discussed above, informants can always construe (more details of the) context themselves. One informant may envisage another context or a more elaborate one than another. Some informants may be very good at quickly construing the relevant context, allowing them to accept questions with interveners more easily. Recall also from Section 2 that a speaker’s associations may influence what a sentence makes salient for them. For example, a person’s beliefs about and experiences with children may influence what the question in (36) makes salient.

Even if a sentence is presented to an informant in a particular context, this context may allow for more than one interpretation, depending on the speaker. An example is the context in (37), adapted from Engdahl (2006).

Does the remark that both children are rather picky establish that there is a specific thing that Anne’s daughter does not eat (at all)? That seems to be open to interpretation. By analogy with the son, there may also be a specific type of food that the daughter does not eat. Nevertheless, the daughter could also just be generally fussy about her food. We call the meaning that would not so easily license the use of a choice function in (37) ‘semi-salient’: the context allows for both an interpretation that makes the choice function recoverable and an alternative interpretation. One can render the context in (37) more explicit so that it leaves no more room for an alternative interpretation (38).

In (38), we (a) added a child, strengthening the sense of analogy, (b) made explicit that each child has the relevant property of not eating a specific thing and (c) made the contextual restriction of the wh-phrase more specific (vegetables instead of food items in general). All these elements increase the salience of a specific type of vegetable that Anne’s daughter does not eat.

In sum, the choice function proposal predicts the observed variation in judgements. Variation is expected because individual speakers may envisage and interpret contexts differently.

5.1 Differences among speakers: the extra-strong presupposition

The view presented in Section 4.1, according to which French wh-in-situ questions involve an extra-strong presupposition, is quite wide-spread in the literature (Boeckx 1999; Boeckx et al. 2001; Boucher 2010; Cheng and Rooryck 2000; Mathieu 2002; Zubizarreta 2003). However, many other authors maintain that French wh-in-situ questions do not involve a stronger presupposition than wh-fronted questions (Adli 2006; Aoun et al. 1981; Beyssade 2006; Hamlaoui 2011; Mathieu 2004; Oiry 2011; Shlonsky 2012; Starke 2001). A third position holds that only some wh-in-situ questions involve an extra-strong presupposition (Baunaz 2005, 2011, 2016; Coveney 1989a,b; Starke 2001; Zimmermann and Onea 2011; in a different framework Myers 2007).

While our proposal aligns with the last position, the ‘Two speaker groups hypothesis’ put forward in this section also accounts for the data variation. Recall that the grammar of Type A speakers only has a contextually licensed choice function to interpret wh-in-situ. Thus, for Type A speakers, French wh-in-situ questions require a particular type of context, which makes the referent for the wh-phrase recoverable (e.g. Chang 1997; Cheng and Rooryck 2000). On the other hand, for Type B speakers, who also have the covert movement option, French wh-in-situ questions can also be used out-of-the-blue (e.g. Adli 2006; Hamlaoui 2011).

Direct evidence for the existence of a group of speakers who can interpret wh-in-situ via covert movement is to be found in Glasbergen-Plas (2021). The author reports on a rating study investigating the felicity of French wh-in-situ long-distance questions in an out-of-the-blue context (as well as indirect wh-in-situ questions and wh-in-situ questions containing an adjunct island).^[19] The contexts used in the study were specifically designed to exclude the possibility of licensing via a contextually bound choice function. An example is presented in (40).

The questions in contexts were rated by monolingual native speakers of French (N = 64, 20–35 years of age) who originated from different regions in France. Questions as in (40) received a mean rating of 4.0 on a scale from 1.0 to 5.0. That is, speakers in a relatively young age group accepted French wh-in-situ questions in a context designed to exclude the use of a choice function. Under the current analysis, this confirms the existence of Type B speakers, speakers who have a second mechanism to interpret French wh-in-situ questions in addition to a choice function. The rating study also suggests that the group of Type B speakers includes speakers in a younger age group (age 20 to 35).^[20]

As indicated above, these conclusions are in line with the findings of Baunaz and Bonan (2023), who show that the proportion of wh-in-situ questions of which the propositional contents is not contextually given increases over time.

5.2 Differences among speakers: the large sentence-final rise

A second area of speaker variation that the ‘Two speaker groups hypothesis’ sheds light on concerns judgements regarding the presence of a large sentence-final rise (i.e. a rise with a large pitch excursion). We first summarise the debate in the literature on this topic before explaining the current proposal’s contribution.

The main significance of the alleged final rise stems from a theoretical paper by Cheng and Rooryck (2000), who propose that wh-in-situ questions are licensed by their rising intonation. A production study by Déprez et al. (2013) provides ‘nuanced support’ (p. 15) for Cheng and Rooryck’s claim, although Tual (2017) could not replicate Déprez et al.’s findings. An opposing view states, however, that French wh-in-situ questions end in a fall (Di Cristo 1998; Mathieu 2002; Starke 2001). Yet a third position holds that a large sentence-final rise is possible, but optional rather than mandatory (Adli 2004, 2006; Wunderli 1978, 1982, 1983; Wunderli and Braselmann 1980). Reinhardt (2019) settles this debate to some extent. In two corpus studies, she shows that while both fronted and in-situ wh-questions occur with a rising as well as a falling final contour, a large sentence-final rise is more frequent in wh-in-situ than in wh-fronted questions.

A production study by Glasbergen-Plas et al. (2021) throws new light on the observed variation. They investigated the prosody of French wh-in-situ questions with either a broad focus or a narrow focus on the wh-word (as well as echo questions). Although it is sometimes assumed that the focus in wh-questions necessarily equals the wh-phrase, this is not the case in all languages (Beyssade 2006; Eckardt 2007; Jacobs 1984, 1991; Ladd 2009; Reich 2002). Wh-questions in French may be prosodically marked for focus (Beyssade 2006; Glasbergen-Plas et al. 2021). Crucially, Glasbergen-Plas et al. show that French wh-in-situ questions with a narrow focus on the wh-word display a large sentence-final rise, while wh-in-situ questions with broad focus do not.^[21] Based on known correlates of focus marking in French, the authors analysed the large sentence-final rise as a correlate of narrow focus marking.^[22] In other words, they show that the large final rise observed in some French wh-in-situ questions is a marker of narrow focus on the wh-word.

A key point to note is that choice function wh-in-situ questions always have a narrow focus on the wh-word. As an illustration of this point, consider a context in which a speaker knows that both Anne and Sophie have booked a restaurant for Thursday evening, to have dinner with their respective families. In this context, the in-situ wh-phrase in (41) is licensed as choice function wh-in-situ, because it has been established in the context that there is a specific restaurant where the addressee made the reservation.

In the described context, the wh-in-situ question has a narrow focus on the word quel ‘which’, as resto ‘restaurant’ and the rest of the sentence are contextually salient. Thus, a context that licenses the use of the contextually bound choice function elicits questions with a narrow focus on the wh-word.

As this large final rise has been shown to be part of the characteristic tune for utterances with a narrow focus on the wh-word (Glasbergen-Plas et al. 2021), the fact that choice function wh-in-situ questions have a narrow focus on the wh-word suggests that these questions tend to be marked by a large sentence-final rise.^[23] Consequently, for Type A speakers, who only have choice function wh-in-situ, French wh-in-situ questions typically display such a rise. Covert movement wh-in-situ questions may occur in various contexts, including out-of-the-blue cases (Glasbergen-Plas 2021; see Section 5.1), hence may also have other focus structures. Given that the contour used in the broad focus condition in Glasbergen-Plas et al. (2021) did not end in a large final rise, such a rise is not a typical feature of French wh-in-situ questions for Type B speakers. Thus, the ‘Two speaker groups hypothesis’ sheds new light on the longstanding debate concerning the alleged presence of a final rise. The results of Glasbergen-Plas et al. (2021) suggest that the authors who typically observed a falling contour (e.g. Mathieu 2002) investigated questions with broad focus.

5.3 Differences among speakers: intervention effects

If a group of French speakers who have both the choice function and covert movement strategies (Type B speakers) are to interpret wh-in-situ questions, the question arises what the predictions are for this group regarding (apparent) intervention effects. The short answer to this is that the predictions for both speaker groups are the same. Type B speakers are not expected to accept a wider range of sentences containing interveners, since an intervener blocks covert movement (cf. (3) in Section 1). The only difference between the two speaker groups regarding interveners is that true intervention effects (i.e. not apparent intervention effects) can only arise for Type B speakers, because they have the covert movement option.

Nonetheless, both Type A and Type B speakers may in some cases accept a wh-in-situ question without an intervener yet fail to accept it when an intervener is added. The argument for why this is so runs parallel to the argument in Section 4.2 about differences between interveners. We argued in Section 4.2 that, when confronted with a sentence without context, it is easier to envisage a context that makes a choice function recoverable for certain interveners, such as plusieurs N ‘several N’, than for others, such as ne que ‘only’ in (33), repeated here as (42).

We furthermore suggest that it is often more difficult to envisage a context that makes the choice function recoverable for a sentence with an intervener than for one without an intervener. For (42) to be interpretable via a choice function, the context must establish a referent for particular circumstances under which Jean only spoke to Suzanne. In contrast, a choice function in the same sentence without the intervener (43) would be interpretable if the context makes salient particular circumstances under which Jean spoke to Suzanne. We suggest that the latter type of context is easier to envisage than the former, because ne que ‘only’ imposes quite a specific restriction on the context (as in the comparison between plusieurs N ‘several N’ and ne que ‘only’ in Section 4.2).

We, therefore, predict that both types of speakers may accept a wh-in-situ question without an intervener. The acceptance goes down when an intervener is added, because speakers may no longer be able to envisage the necessary context.

5.4 Differences among speakers: summary

The properties of wh-in-situ questions for the two groups of speakers are summarised in Table 1.

The first concerns the aforementioned ‘extra-strong-presupposition’. Under our analysis, authors who argue for the presence of the presupposition (e.g. Boeckx 1999; Chang 1997; Cheng and Rooryck 2000) rely on judgements of Type A speakers while authors who argue against such a presupposition (e.g. Adli 2006; Mathieu 2004; Shlonsky 2012) rely on those of Type B speakers. In the absence of an intervener, Type A speakers still need an appropriate context to make the choice function recoverable, while Type B speakers do not, because they can rely on covert movement.

The second debate concerns the large sentence-final rise (Cheng and Rooryck 2000; Déprez et al. 2013). Under our analysis, choice function wh-in-situ questions always have a narrow focus on the wh-word. A large final rise is a prosodic correlate of this focus structure (Glasbergen-Plas et al. 2021). In other words, for Type A speakers, wh-in-situ questions always display a large final rise, while for Type B speakers, various prosodic tunes are possible depending on the context, including a falling contour.

Lastly, we have shown that it is easier for speakers to contextually establish a unique referent for the wh-phrase for certain interveners than for others. This explains the observed variation in judgements between different interveners (e.g. Baunaz 2005, 2011). We also illustrated how different speakers may envisage and interpret contexts differently. This sheds light on the observed speaker variation regarding intervention effects (e.g. Adli 2006; Mathieu 2002). It is important to realise that interveners do not differentiate between the two groups of speakers: both groups need to interpret a wh-in-situ question containing an intervener via a choice function, as an intervener blocks covert moment.

There seems to be a tendency for older literature to present data that, under the current analysis, represent a Type A grammar (only a choice function) (e.g. Boeckx et al. 2001; Chang 1997; Cheng and Rooryck 2000), whereas later literature is more often consistent with a Type B grammar (also covert movement) (e.g. Baunaz 2011; Reinhardt 2019). Even though this is far from a clear correlation across the relevant publications, we suggest that the picture is consistent with the idea that the language is changing: the grammar of Type A speakers may reflect an earlier variety of colloquial French. This variety would be a wh-fronting language that only allows for contextually restricted wh-in-situ. Possibly, contextually restricted wh-in-situ was gradually used more freely, for instance in the case of ‘semi-salience’ (see Section 4.3). Thus, a Type B grammar would emerge, in which covert movement is permitted to resolve in-situ wh-phrases, leading to felicitous use of out-of-the-blue wh-in-situ questions by a subset of speakers. This scenario is in accordance with observations by Larrivée (2019) and the corpus data presented in Baunaz and Bonan (2023).

6.1 Comparison with Baunaz (2005, 2011, 2016), building on Starke (2001)

We now compare our account to the proposal by Baunaz (2005, 2011, 2016), which builds on Starke (2001) (henceforth: B/S). B/S propose a hierarchy of nanosyntactic features for both in-situ wh-phrases and interveners. As Baunaz (2016: 164) states, a wh-phrase ‘that has something more than a potential intervener is free to move, whereas a [wh-phrase] that has either something less, or that has a similar feature composition as a potential intervener, will be blocked’. Specifically, Baunaz distinguishes between ‘specific’, ‘partitive’ and ‘non-presuppositional’ wh-phrases and interveners, where ‘specific’ has ‘something more’ than ‘partitive’ and ‘partitive’ has ‘something more’ than ‘non-presuppositional’.^[24] Therefore, a ‘partitive’ wh-phrase would be able to cross a ‘non-presuppositional’ intervener, but not a ‘specific’ one.

The example in (44) illustrates for Baunaz the extraction of a partitive wh-phrase.

In the original example, the wh-phrase is written as singular (quelle machine), and Baunaz assumes that this is a partitive wh-phrase: there is a pre-established set of machines (the ones that Claire did not use) and the singular wh-phrase picks out one of these machines. Partitive readings are not expected to be possible under the current proposal: as one can conceive of different subsets from the mentioned set, the reference of the wh-phrase is not determined, and thus not recoverable from the context.

However, as the plural markers are not pronounced, the question in (44) does not specify whether quelle(s) machine(s) is a singular or a plural form. In the given context, it is likely that the coach wants to know which machine(s) Claire did not use, as this is the information they need for preparing the program for the next session. If Claire did not use the rowing machine, this machine might for example be referred to as ‘the rowing machine’, ‘that big one’ or ‘the last one in the circle’. On the other hand, if Claire did not use the rowing machine nor the cycling bike, these two form a plural individual that can for instance be referred to as ‘the rowing machine and the cycling bike’, ‘the last two in the circle’ or ‘the ones she dislikes most’. Whichever description is used in the answer to the wh-question, any congruent answer refers to the referent that is identified by the context, the ‘one or more machines that Claire did not use’. Hence, the referent is uniquely identifiable, and the question might be paraphrased by the identity question in (45), as in the case of the examples with specific wh-phrases (in Baunaz terms) discussed in Section 2.

We conclude that (44) is not a counterexample to our proposal and suggest that there is no need to assume the category ‘Partitive’ in Baunaz’s hierarchy.

The current proposal has a number of important advantages over B/S’s account. First, it accounts for more data. In contrast to B/S, our account explains the observed variation in judgements among sentences containing the same intervener. Likewise, in contrast to B/S, it explains why a subset of speakers (Type A speakers in our terms) only accept wh-in-situ in a particular type of context, regardless of whether the question contains an intervener or not. Second, the current account explains the contextual effects found for wh-in-situ without assuming the presence of a [+specific] feature. The observed effects are accounted for by a felicity condition on the interpretation of the choice function, which needs to be recoverable for the interlocutor. Since the mechanism of a choice function is directly related to in-situness, the semantic effects are expected to occur for in-situ wh-phrases only. In B/S’z account, it is not clear why in-situ but not fronted wh-phrases exhibit the proposed feature hierarchy. Cross-linguistically, specificity correlates with movement. Both in Turkish (Enç 1991) and in Germanic languages (De Hoop 1992, among others), specific noun phrases typically undergo scrambling or object shift, while non-specific noun phrases do not. The presence of a feature [+specific], as proposed by B/S, is therefore not expected to correlate with in-situness. Finally, given the ‘Intervention effects avoidance generalisation’ as stated in (16), we expect that the referent for the wh-phrase is recoverable in the context. Within the proposal of B/S, this descriptive generalisation should be an effect of the specificity of the wh-phrase, but it is not clear why specificity would have such an effect, given that specific noun phrases can be used in sentences with different types of focus-ground articulations.

6.2 Cross-linguistic look-ahead

In this last section, we turn to the cross-linguistic picture of wh-in-situ in wh-fronting languages. In what languages would choice function single wh-in-situ questions be expected to emerge? We expect that a wh-fronting language that has multiple wh-questions (such as English) can develop contextually restricted single wh-in-situ. Multiple wh-questions, with one fronted and one in-situ wh-expression, have two possible readings: single pair and pair-list. According to Dayal (2002), both choice functions and covert wh-movement are needed to derive these two readings. In multiple wh-questions, the choice function variable is unselectively bound by the existential quantifier introduced by the fronted wh-phrase. In the absence of a fronted wh-phrase, however, there is no binder within the sentence. We assume that in that case, contextual binding may become an option resulting in contextually restricted wh-in-situ.

This expectation seems to be borne out. Several languages with multiple wh-questions have been reported to allow a restricted use of wh-in-situ, which is only acceptable in a specific set of contexts (Biezma 2018; Bobaljik and Wurmbrand 2015; Bolinger 1978; Ginzburg and Sag 2000; Jiménez 1997; Poschmann 2015; Roussou et al. 2014). These languages include English, German, Spanish, (Brazilian) Portuguese and Modern Greek.

We suggest that the generalisation presented in Section 2 for French also goes a long way towards analysing the contextual restrictions of wh-in-situ questions in English and German, which are described in detail in the literature. In (46) and (48), the referent for the wh-phrase is already established is the context, as illustrated by the identity questions in (47) and (49), respectively.

As with contextually restricted wh-in-situ in French, the acceptability of examples such as (46) and (48) seems to vary, with certain speakers being able to accept wh-in-situ more easily than others. Hence, the choice function account proposed for French may also be a promising direction of research for wh-in-situ in languages such as English and German. If this contextually restricted wh-in-situ is gradually used more freely, wh-in-situ in such languages might stand at the beginning of a similar language change as the one we proposed for French in Section 5. Future research will have to show whether the choice function analysis developed in this paper can be extended to other languages and whether the further development towards a system allowing covert movement in single wh-questions can also be observed for other languages than French.