Q-movement Linguistics

Q-movement 1. The proposal The proposal I will be putting forward in this talk is that there is a particle appearing in questions which undergoes syntactic movement, even in languages which do not show any obligatory movement of the question words themselves. I will primarily concentrate on questions in Japanese. As an illustration of the proposal, consider the pair of Japanese sentences given in (1) on the handout. The first, (1a), is a declarative sentence which means ‘someone bought a book’. The second, (1b), is a question which means ‘who bought a book?’ Importantly, notice that the only difference between these two sentences, apart from their intonation contour, is the order in which the morpheme -ka appears. In (1a), -ka appears as part of dareka ‘someone’. In (1b), -ka appears at the end of the utterance. (1) a. dareka-ga hon-o kaimasita. someone-NOM book-ACC bought.POLITE ‘Someone bought a book.’ b. dare-ga hon-o kaimasita ka? who-NOM book-ACC bought.POLITE Q ‘Who bought a book?’ Japanese The main proposal of this talk is illustrated in (2). It is that -ka in (1b) appears at the end of the question as a result of syntactic movement, carrying it from a position next to the question word dare ‘who’ to its observed position at the edge of the clause. (2) “Q-movement”— In Japanese questions, -ka reaches the clause periphery via syntactic movement. dare-tka-ga hon-o kaimasita ka ? 1 " z------------------m =(1b) The rest of this talk will be devoted to finding evidence for “Q-movement” of this kind. We will start by looking at two diagnostics which are sensitive to syntactic movement, the first being “intervention” or “relativized minimality” effects, and the second being island effects. Then, we will look at question formation in an unrelated—but structurally quite similar—language, and then at question formation in historically earlier stages of Japanese. 1 Plan: • Review movement diagnostics: Intervention effects Island effects Interactions • Question formation in Sinhala (crosslinguistic support) • Question formation in Premodern Japanese (historical support) 2. Intervention effects The first kind of evidence we will look at comes from “intervention effects.” Assuming that syntactic movement is constrained to be as “short” as possible, an “intervention effect” occurs if a movement which is required for well-formedness is longer than another movement which could have occurred instead. The idea is illustrated in (3); here, X cannot move over Y if Y was eligible for movement, since moving Y would result in a shorter move. (3) 3 X … 3 " 1 Y … 1 3 z--=-- tX … Y intervenes for movement of X if Y is equally eligible to move, since Y is closer to the target of movement than X was. Taking movement to be driven by a requirement of the target of movement, we can encode this as the principle Attract Closest, stated in (4). (4) Attract Closest Movement is motivated by the need to check a feature. Only the closest element with the relevant feature is eligible for movement. With respect to Japanese questions, this means that Q-movement is motivated by some requirement of a complementizer or complementizer-like element, and that the closest -ka to that complementizer moves to satisfy the requirement. With this theory of movement in mind, we can now test to see if Q-movement exhibits the intervention effects we would expect to find if -ka in questions really moves from inside the clause. First, we locate other instances of -ka, like those in the sentences in (5). In (5a), we see that -ka can be used to mean‘or’, as in John-ka Bill ‘John or Bill’. In (5b), we see the example from before, where -ka is used as part of the indefinite dareka, which means ‘someone’. Supposing that this -ka is the same morpheme as the -ka which is found at the end of questions, we can now test what happens if one of these other kinds of -ka is placed along the hypothesized path of Q-movement. 2 Finding intervenors for -ka-movement: Independent instances of -ka. (5) a. John-ka Bill-ga hon-o kaimasita. John-or Bill-NOM book-ACC bought.POLITE ‘John or Bill bought a book.’ b. dareka-ga hon-o kaimasita. someone-NOM book-ACC bought.POLITE ‘Someone bought a book.’ disjunctive indefinite forming We will assume a clause structure of Japanese roughly as diagrammed in (6), abstracting away from various irrelevant details. What is important here is that the place where -ka moves to in questions is hierarchically above the subject. This means that if -ka were to move from next to a question word in object position, the path of movement would cross the subject. Therefore, we can test for intervention effects with object question words and intervenors—like the independent uses of -ka illustrated in (5)—as part of the subject. (6) CP 3 IP complementizer(s) 3 VP tense 3 subject 3 object verb Q moves to complementizer position Q moves from next to question word Moving from object position crosses the subject position. To test: Object question word, intervening -ka inside subject. Hoji, in his 1985 thesis, tested exactly this configuration (although from a somewhat different theoretical perspective). He observed that when either disjunctive -ka or the indefinite dareka appear in wh-questions, the word order is restricted. In general, Japanese word order is fairly free, but in just these cases, the question word must precede the other instance of -ka. So, (7a) is ill-formed because John-ka Bill ‘John or Bill’ precedes nani ‘what’—that is, the other -ka, here the one meaning ‘or’, stands between the object position and the complementizer position. If the order of subject and object is reversed, as it is in (7b), the question is fine. Let us assume (as is standard) that this “scrambling” in (7b) that allows the object to precede the subject is movement of the object to a position higher in the tree (for example, adjunction to IP). The fact that (7b) is well-formed indicates that -ka can move from this higher position to the complementizer without having to cross the other -ka that is part of the subject. 3 Independent -ka cannot be on the path of Q-movement (data noticed by Hoji 1985) (7) a. ?* John-ka Bill-ga nani-o kaimasita ka? John-or Bill-NOM what-ACC bought.POLITE Q (‘What did John or Bill buy?’) Comp -ka John-ka Bill-ga nani-o tka " 1 z-----=----------m b. kaimasita nani-oi John-ka Bill-ga ti kaimasita ka? what-ACC John-or Bill-NOM bought.POLITE Q ‘What did John or Bill buy?’ a----------l ? 1 ti Comp -ka [ nani-o tka ]i John-ka Bill-ga " 1 z--------m (8) (9) kaimasita a. ?* dareka-ga nani-o kaimasita ka? someone-NOM what-ACC bought.POLITE Q (‘What did someone buy?’) b. nani-oi dareka-ga ti kaimasita ka? what-ACC someone-NOM bought.POLITE Q ‘What did someone buy?’ a. dare-ga sake-ka beeiru-o nomimasita ka? who-NOM sake-or beer-ACC drank.POLITE Q ‘Who drank sake or beer?’ b. ?* [sake-ka beeiru-o]i dare-ga ti nomasita ka? sake-or beer-ACC who-NOM drank.POLITE Q (‘Who drank sake or beer?’) This is exactly what Attract Closest and the hypothesis of Q-movement led us to expect. If Q-movement starts from object position, it crosses the subject—and if, as in (7a) and (8a), the subject contains an independent instance of -ka, this movement will violate Attract Closest, and will be ill-formed. The intervention effects reviewed in this section therefore support the Q-movement hypothesis. 4 3. Island effects (or a lack thereof?) Now, let us turn to look at the effect of “islands”, which serve as a second diagnostic for movement. Movement generally cannot cross island boundaries, which include relative clauses and adverbial adjuncts. This is taken to be the source of the ill-formedness of the English questions in (10)—the question word what has been moved from a position inside an island to a position outside the island. (10) a. b. * Whati did John meet [island the man who wrote ti ]? * Whati did Mary leave [island after John bought ti ]? So let us suppose that movement can never cross island boundaries. In the context of the claim I am making in this talk, that Japanese -ka moves to its surface position from inside the clause, we would expect that putting an island boundary along the path of movement should render a question ill-formed, much like in (10a). However, at first sight, this seems not to be the case. For example, Japanese has no difficulty with questions that have question words inside islands, as shown in (11). So, in (11a), the question word nani ‘what’ is inside a relative clause, but the question is well formed. Japanese allows question words inside islands (11) a. Taroo-ga [island Hanako-ni nani-o ageta hito-ni ] aimasita ka? Taroo-NOM Hanako-DAT what-ACC gave man-DAT met.POLITE Q ‘What did Taro meet the man that gave t to Hanako?’ b. yomu maeni ] dekakemasita ka? Taroo-ga [island Hanako-ga nani-o Taroo-NOM Hanako-NOM what-ACC read before left.POLITE Q ‘What did Taro leave before Hanako read t?’ Given our assumptions, we can only conclude that in (11), Q-movement does not cross the island boundary—that is, that -ka does not move to its surface position from inside the island. Rather, what movement there is must have taken place from outside the island. This scenario is illustrated in (12); -ka moves from outside the island to its clause-final position. (12) [island … nani … ] … tka … ka ? 1 " z--m This state of affairs verges on making islands useless as a diagnostic for movement, since if -ka can always move from outside the island we will never see island effects. However, it turns out that there is a way to tell where Q-movement originated, with the help of the emphatic particle ittai. In a simple question like (13), ittai before the question word nani adds an emphatic meaning, something like English ‘what in the world?’. 5 An example of the emphatic particle ittai (13) John-ga ittai nani-o kaimasita ka? John-NOM ittai what-ACC bought.POLITE Q ‘What in the world did John buy?’ The ittai particle appears to have a close relation to the question particle -ka. It is generally possible to signal a question in Japanese solely through intonation, without a final -ka, as in (14a). However, a question like (14b) with ittai sounds much worse when it lacks -ka. ittai requires presence of -ka—without ittai, -ka can sometimes be dropped. (14) a. Taroo-ga nani-o tabeta? Taroo-NOM what-ACC ate ‘What did Taro eat?’ b. ?? Taroo-ga ittai nani-o tabeta? Taroo-NOM ittai what-ACC ate (‘What in the world did Taro eat?’) The position of ittai by the question word (where Q-movement is hypothesized to start) and its apparent connection to the ‘Q’ particle -ka suggests a diagnostic use to which ittai can be put. Specifically, ittai seems to “mark” the place from which Q-movement occurred—ittai seems to get “left behind” when -ka moves. This means that we can use ittai to test directly what happens if Q-movement crosses an island boundary. Consider the questions in (15), which are the questions from (11) but with ittai added next to the question word inside the island. Since ittai gets “left behind” when -ka moves, this configuration implies that Q-movement must have crossed an island boundary (starting from inside, ending up outside) in (15), and so these questions are ill-formed, as we would expect. Hypothesis: ittai gets left behind when -ka moves ittai inside an island ⇒ -ka moved out of the island ∴ ill-formed (15) a. * Taroo-ga [island Hanako-ni ittai nani-o ageta hito-ni ] Taroo-NOM Hanako-DAT ittai what-ACC gave man-DAT aimasita ka? met.POLITE Q (‘What in the world did Taro meet the man that gave t to Hanako?’) 6 b. * Taroo-ga [island Hanako-ga ittai nani-o yomu maeni ] Taroo-NOM Hanako-NOM ittai what-ACC read before dekakemasita ka? left.POLITE Q (‘What did Taro leave before Hanako read t?’) Interestingly enough, if ittai appears just outside the island, as in (16), the question is wellformed. The structure of (16) is presumably just like (11a), where -ka moves from a position outside the island to its surface position at the end of the clause. ittai just outside an island, fine (-ka moves from outside the island) (16) Taroo-ga ittai [island Hanako-ni nani-o ageta hito-ni ] Taroo-NOM ittai Hanako-DAT what-ACC gave man-DAT aimasita ka? met.POLITE Q ‘What in the world did Taro meet the man that gave t to Hanako?’ Wrapping up what we’ve seen in this section, Japanese allows question words inside islands, indicating that in such cases -ka moves from a position outside the island to its surface position. Using ittai to mark the launching site of -ka, we also saw that a question in which -ka is forced to move across an island boundary is ill-formed, as expected. 4. Islands and intervention effects There is another diagnostic which is sensitive to the path of Q-movement that we can use to corroborate the conclusion that -ka moves from outside islands containing a question word. This diagnostic is the intervention effect discussed back in section 2. The idea is this: Recall that “intervention effects” occur when another -ka is along the path of Q-movement. Now, when a question word is inside an island, we hypothesized that Q-movement starts outside the island. This means that Q-movement should not cross over anything inside the island, and so there should be no intervention effects inside an island. Put at its most counterintuitive, a question which is ill-formed due to an intervention effect should be improved by embedding it in an island. Perhaps surprisingly, this prediction is borne out. Recall the intervention case from before, repeated as (17). (17a) is ill-formed because the other -ka—here the particle meaning ‘or’—is between nani ‘what’ and the complementizer level of the clause structure. (17b) is fine because nani (and thus the place from which Q-movement occurs) is scrambled to a position above the other -ka. Idea: • Independent -ka blocks Q-movement (“intervention effect”) • Q-movement takes place from outside of islands. • Q-movement should be insensitive to intervenors inside an island 7 Recall: Paradigm showing intervention effect. (17) a. ?* John-ka Bill-ga nani-o kaimasita ka? John-or Bill-NOM what-ACC bought.POLITE Q (‘What did John or Bill buy?’) b. John-ka Bill-ga ti kaimasita ka? nani-oi what-ACC John-or Bill-NOM bought.POLITE Q ‘What did John or Bill buy?’ Now, consider (18). In (18), we have embedded the questions from (17) inside an adverbial island. Counterintuitively—but as predicted—the question improves. In particular, (18a) is a well-formed question, despite the fact that John-ka Bill precedes the question word nani. Comparing (18a) to (18b) shows that the relative order between the disjunction and the question word does not matter when they are inside an island—unlike in (17). This is simply because the path of Q-movement never crosses the subject inside the island, regardless of its order with respect to the question word. Embed “intervention effect” in an island—becomes well-formed (order becomes free) (18) a. Mary-wa [island John-ka Bill-ga nani-o katta ato de ] Mary-TOP John-or Bill-NOM what-ACC bought after dekakemasita ka? left.POLITE Q ‘Mary left after John or Bill bought what?’ b. Mary-wa [island nani-oi John-ka Bill-ga ti Mary-TOP what-ACC John-or Bill-NOM dekakemasita ka? left.POLITE Q ‘Mary left after John or Bill bought what?’ katta ato de ] bought after 5. Question formation in Sinhala Sinhala: Indo-European, spoken in Sri Lanka, SOV with scrambling, wh-in-situ. We have now developed the proposal in considerable detail based only on the facts from Japanese. At this point, let us switch gears and consider question formation in a different language, which although unrelated to Japanese will provide support for the hypotheses we have been developing. The language we will consider is Sinhala, an Indo-European language spoken in Sri Lanka. Sinhala shares with Japanese the basic word order SubjectObject-Verb, and the ability to “scramble” the surface word order. In Sinhala as in Japanese, there is no obligatory movement of question words. 8 Compare the declarative sentence meaning ‘Chitra bought that book’ in (19) with the related question ‘What did Chitra buy?’ in (20). There are three features to notice about the question. First, mokak ‘what’ in (20) appears in the same place as the object in its declarative counterpart. Second, the question word is followed by a particle d\, which I have glossed as ‘Q’. Third, the verb appears in a special form, as indicated by the final vowel (which has changed from -a to -e). (19) (20) Chitra ee pot\ gatta. Chitra that book bought ‘Chitra bought that book.’ Sinhala Question word remains in situ. Question word followed by Q. Verb takes special “-e” form. Chitra mokak d\ gatte? Chitra what Q bought-E ‘What did Chitra buy?’ The proposal I would like to defend here is that the particle d\ in the Sinhala question in (20) directly corresponds to the particle -ka in Japanese questions discussed previously—with the important difference being that in Sinhala we see the particle in its pre-movement position, whereas in Japanese we see the particle in its post-movement position. To increase the plausibility of identifying Japanese -ka with Sinhala d\, let me point out two parallels between them. Recall from the beginning of this talk that in Japanese, a question word with -ka after it—like dareka in (21)—is interpreted as an indefinite, here ‘someone’, built of dare ‘who’ and -ka. Reasons to believe Sinhala d\ corresponds to Japanese -ka: (a) Wh+Q = Indefinite (21) dareka-ga hon-o kaimasita. someone-NOM book-ACC bought.POLITE ‘Someone bought a book.’ Japanese Sinhala d\ has the same property. When it appears after a question word in a declarative sentence like (22), it also yields an indefinite interpretation. Here, mokak d\, built of mokak ‘what’ and d\, means ‘something’. (22) Chitra mokak d\ gatta. Chitra what Q bought ‘Chitra bought something.’ Sinhala A second point of parallelism between Japanese -ka and Sinhala d\ is that both are used in yes-no questions. In each language, a yes-no question can be formed by appending the corresponding particle to a declarative sentence. So, in Japanese, the question ‘Did Taro buy a book?’ in (23a) is formed by appending -ka to the sentence in (23b) which means ‘Taro bought a book.’ In Sinhala, the question ‘Did Chitra buy that book?’ in (24a) is formed by appending d\ to the sentence in (24b) which means ‘Chitra bought that book.’ 9 Reasons to believe Sinhala d\ corresponds to Japanese -ka: (b) declarative+Q = y/n ques. (23) (24) a. Taroo-ga hon-o kaimasita. Taroo-NOM book-ACC bought.POLITE ‘Taro bought a book.’ b. Taroo-ga hon-o kaimasita ka? Taroo-NOM book-ACC bought.POLITE Q ‘Did Taro buy a book?’ a. Chitra ee pot\ gatta. Chitra that book bought ‘Chitra bought that book.’ b. Chitra ee pot\ gatta d\? Chitra that book bought Q ‘Did Chitra buy that book?’ Japanese Sinhala Accepting that Japanese -ka corresponds to Sinhala d\, with the hypothesis that the surface position of Sinhala d\ reflects the pre-movement position of Japanese -ka, the next logical step is to look at questions involving islands. Recall that our hypothesis based on Japanese was that Q-movement occurs from outside the island in such questions. We were led to this by the fact that question words are well-formed inside islands in Japanese. In Sinhala, we also find that question words are allowed inside islands—but only if d\ appears outside the island. So, for example, (25a) is ill-formed because d\ appears next to the question word inside the island. (25b), on the other hand, is fine because d\ appears just outside the island. This is exactly the structure we hypothesized for Japanese, but in Sinhala we can actually see the pre-movement position of Q on the surface. Sinhala: Question words allowed in islands—but only if d\ is outside. (25) a. b. * [island Chitra mon\wa d\ kan\ kot.\ ] Ranjit pudum\ unee ? Chitra what Q ate when Ranjit surprise became-E (‘Ranjit was surprised when Chitra ate what?’) [island Chitra mon\wa kan\ kot.\ ] d\ Ranjit pudum\ unee ? Chitra what ate when Q Ranjit surprise became-E ‘Ranjit was surprised when Chitra ate what?’ Tying this all together, the pattern in (25) suggests that, although d\ appears in its “pre-movement” position, there still is a movement relation between the surface position of d\ and the edge of the clause. That is, the reason that d\ cannot be separated from the edge of the clause by an island boundary is that d\ moves, just like -ka in Japanese—except that while Japanese -ka moves overtly, Sinhala d\ moves “covertly.” This is where the special verb form that appears in Sinhala questions comes in. Recall that in questions, the verb appears in a special form, ending in “-e”. Our assumption 10 about movement has been that it is motivated by the need to “check a feature” or satisfy a requirement at the landing site of movement. So, in questions, there is some requirement of the complementizer system that movement of Q can satisfy. In Sinhala, the special verbal suffix -e can be viewed as an overt morphological realization of this “unchecked feature”—it indicates that there will be a “covert movement” to satisfy the requirement it reflects. (26) The Sinhala -e verbal form is a morphological reflection of an unchecked feature in the complementizer system (the feature which drives movement of d\). 6. Premodern Japanese kakari-musubi It is worth pointing out that we can find, in the history of Japanese itself, evidence of a stage which is striking similar to modern Sinhala. That is, Japanese seems to have historically changed from a language with “covert Q-movement” to the present-day language with “overt Q-movement.” In Premodern Japanese, questions often took a form referred to as kakari-musubi, where -ka would surface next to a question word and the verb would surface in a special (attributive) form. An example is given in (27). The question particle is the familiar -ka, here attached to the question word tare ‘who’. The verbal suffix glossed as ‘M’ (for musubi) appears in these questions, quite parallel to the -e suffix in Sinhala. Premodern Japanese: Q next to question words, verb takes special (musubi) form. (27) tare-ka mata hanatatibana-ni omoi-idemu. Premodern Japanese who-Q again flower.orange-DAT remember-M ‘Who will again remember (me) at the time of the mandarin orange flower?’ (Shin Kokin Wakashð [1205]:3) Although we of course do not have access to native speaker judgments of Premodern Japanese, there are some recorded cases in which question words appear inside islands—and in these cases, -ka appears not next to the question word inside the island but rather just outside the island. An example showing this is given in (28). Again, this is just like the behavior of d\ in modern Sinhala. As in Sinhala, Q appears just outside of islands. (28) [island ika yoo naru kokorozasi aramu hito-ni ]-ka awamu to obosu. how kind is love have person-DAT-Q wed that think-M ‘[What kind of love]i do you think you would want to marry a person that has ti?’ (Taketori Monogatari [c. 900]) 11 7. Summary and conclusion Let me end at this point with a reiteration of the proposal. In Japanese and Sinhala whquestions, there is a question particle which moves from a clause-internal position to a clause-peripheral position. This movement begins from a position next to the question word except if that question word is in an island, in which case the movement begins just outside the island. (29) “Q-movement”— In wh-questions (Japanese, Sinhala), ‘Q’ moves. wh-word-tQ [island … wh-word … ] … Q? 1 " z----m tQ … Q? 1 " z--m 12