ROOT MERGER IN CHINESE COMPOUNDS*

Niina Ning Zhang

Abstract. Working on the computation of roots in Chinese compounds, this paper presents six major dierences between the merger of roots in compound formation and the merger of elements in phrase structure formation: the existence of exocentric structures, the freedom of projectivity, the disappearance of subcategorization, the double licensing of formal features, the eect of Lexical Integrity in movement, and such an eect in pronominalization. All of these dierences are accounted for by the hypothesis that roots do not have syntactic features. Moreover, this paper studies a case of merger level underspecication in languages that allow nouns and verbs to occur without any inection: in the absence of a syntactic context, an isolated string of two elements can be ambiguous between the root-merger that eventually gives rise to a verb and the merger of a verb and its object that gives rise to a VP.

1. Introduction In syntactic approaches to morphology (e.g. Halle & Marantz 1993, 1997, Harley & Noyer 2003, Embick & Noyer 2005, Borer 2005a, b), morphemes are divided into abstract or functional morphemes and root or lexical morphemes (substantive listemes). The former are composed exclusively of non-phonological features, such as [PAST], [PLURAL], or features that make up articles, pronouns, and classiers (see Borer 2005a: 96), whereas the latter include items such as BEE or SWIM, which are composed of phonological and conceptual features. Roots are language specic combinations of sound and meaning. Importantly, in such syntactic approaches, it is assumed that roots do not contain or possess grammatical (syntactic or syntactic-semantic) features (or properties). This assumption predicts that an element that expresses a certain concept may project in dierent categories, depending on the formal requirements of the syntactic conguration. The assumption is in contrast to the idea that the grammatical category of a word is dependent on the meaning expressed by it, an idea that can be traced back as far as Aristotles Poetics. The empirical advantages of the assumption over the latter idea have been argued for in studies such as bregas Marantz (1997, 1999), Barner & Bale (2002), Arad (2003), Fa (2005), Borer (2005a, b), and Harley (2005). In complement to the research that shows how lexical morphemes get their syntactic features from the syntactic contexts, this paper demonstrates how the absence
* I am grateful to a reviewer and the editor of Studia Linguistica and Tzong-Hong Jonah Lin for comments, and to James Myers for both comments and English advice. All remaining errors are mine.
Studia Linguistica 61(2) 2007, pp. 170184. The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007. Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA

Root merger in Chinese compounds 171 of syntactic features in lexical morphemes accounts for their special ways of computation. The goal of this paper is to show the descriptive and explanatory power of the above assumption from a dierent perspective. While most syntactic approaches to morphology focus on the computation of abstract morphemes, and the interactions or relationship between abstract morphemes and root morphemes (e.g. Arad 2003, bregas 2005, Irwin 2006, Myers to appear), few are about the merger Fa of one root morpheme with another root morpheme, including the type of merger that derives categorically exocentric compounds (e.g. the English nouns must-have and wanna-be). One comprehensive study of the syntax of compounds is Josefssons (1995, 1998) work on Swedish. She claims that given the idea that Merge is free and operating at no cost, compounding with a verb and any type of segments should be allowed (Josefsson 1998: 69). The empirical issue to be discussed in this paper is the computation of various roots in Chinese compounds, including categorically exocentric compounds.1 The computation violates several well-recognized syntactic constraints, and is restricted in some unexpected ways. The apparent abnormality can be given a unied account by the assumption that roots do not have syntactic features. If roots have no syntactic features, their merger, as stated in Josefssons above claim, is free. Moreover, in addition to the categorial underspecication revealed by the syntactic approaches to roots (e.g. red can be either an adjective or a noun), another type of underspecication will be explored in this paper: the merger level underspecication. Without a context, an isolated string of two elements can have two possible syntactic derivations: it can be either the result of root-merger or the result of the merger of two elements that have syntactic features. For instance, the string guan xin carry heart can be a verb after two steps of merger: the merger of the two roots, guan and xin, which have no category features, and the merger of the root complex with a functional category, that provides the [V] feature. The string can also be a VP after the verb guan is merged with the NP xin. In both cases, the string means be worried about. The paper is organized as follows. Section 2 presents the abnormality of compounding in Chinese, section 3 is my account for the abnormality in the assumption that roots have no syntactic features, and section 4 uses the same assumption to explain the structural properties of a string like guan xin, which behaves like a word in one context but like a phrase in another context. Finally, section 5 concludes the paper.
1 In this paper, we do not consider resultative and causative V-V compounds, since we have not excluded the possibility that they are derived by head movement (see Matushansky 2006 for syntactic head movement).

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

172 Niina Ning Zhang 2. The abnormality of compounding in Chinese 2.1. Exocentric compounds When one element is syntactically merged with another element, the category label for the resultant element must be identical to one or the other of the two elements (Chomsky 1994: 10). In other words, all syntactic complexes are categorically endocentric. However, it is wellknown that compounds can be exocentric, i.e., there is no need for the category of the mother to be identical to either of its daughters. Patterns like the following are so productive to make Huang (1998) claim that Chinese is a headless language (all compounds under consideration are in bold print in the examples in this paper).2 (1) a. zhe zhang zhuozi de da-xiao this cl table mod big-small the size of this table b. Wo hao-dai zhao-le fen gongzuo. I good-bad nd-prf cl job I have found a job anyhow. (2) a. yi ge kai-guan one cl open-close a switch (e.g. a power switch) b. yi ge hen bao-shou de ren one cl very keep-defend mod person a very conservative person (3) a. Wo yao wu-se yi ge zhu-shou. I want thing-color one cl assist-hand I want to look for an assistant b. hen mao-dun very spear-shield very contradictory A-A ) N

A-A ) Adv

V-V ) N

V-V ) A

N-N ) V

N-N ) A

Two adjective-like roots form a nominal in (1a) and an adverb in (1b). Two verb-like roots form a noun in (2a) and an adjective in (2b). Two noun-like roots form a verb in (3a) and an adjective in (3b). In the following (4a), a verb-like root and a noun-like root form an adjective. In (4b), an adjective-like root and a noun-like root form a verb. Finally, in (4c), an adjective-like root and a verb-like root form a noun.

2 The abbreviations used in the Chinese examples are: exp: experience aspect, prf: perfect aspect, prog: progressive aspect, cl: classier, prt: sentence-nal particle, q: question, mod: modication.

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 173 (4) a. yi jian hen kai-xin de shi one cl very open-heart mod thing a very happy thing b. Ta yizhi zai nali pin-zui. he continuously at there poor-mouth He is talking garrulously over there. c. yi ben xiao-shuo one cl small-say a novel V-N ) A

A-N ) V

A-V ) N

All of these compounds are categorically exocentric. In the syntactic approaches to root merger such as Roeper et al. (2001) and Mukai (2004), exocentric compounds are excluded in their consideration. However, such data challenge any syntactic analysis of root merger. 2.2. The freedom of projectivity If a transitive verb is merged with a nominal element (an instance of so-called set-merge in Chomsky 2000: 133), it is the V feature of the verb, rather than the N feature of the nominal, that projects. As expected, in (5a), dao road looks like the object of zhi know, and the whole string zhi-dao is a verb. However, we also see the opposite situation in compounds like (5b). In this example, ji self seems to be the object of zhi know, however, the whole string zhi-ji is a noun. Compound nouns like ling-shi lead-event ) consul, hua-shi dissolve-stone ) fossil and zhushou assist-hand ) assistant pattern with (5b). (5) a. Wo zhi-dao. I know-road I know. b. wo de zhi-ji I mod know-self my condent or close friend V-N ) V

V-N ) N

Moreover, when an adjective modies a noun (called pair-merge in Chomsky 2000: 133), the whole complex should be a nominal, which cannot be selected by a degree word. As expected, in (6a), da big modies mi rice, and the whole string da-mi rice is a noun, which cannot be selected by the degree word hen very. In (6b), however, although da big modies dan gall bladder, da-dan brave, as an adjective, is selected by the degree word hen very. (6) a. yi dai (*hen) da-mi A-N ) N one bag very big-rice a bag of rice b. yi ge hen da-dan de ren A-N ) A one cl very big-gall.bladder mod person a very brave person
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

174 Niina Ning Zhang The examples in these two subsections demonstrate the freedom of root merger in Chinese with respect to categories. Trying to nd some headedness systems in Chinese compounding, Packard (2000: 39) proposes the following principle: (7) a. (bisyllabic) nouns have nominal constituents on the right b. (bisyllabic) verbs have verbal constituents on the left Examples such as (6a) are compatible with (7a), and examples such as (5a) are compatible to (7b). However, as recognized by Packard (2000: 127), there are quite a lot of counterexamples to (7). In addition to the examples discussed above, we can see that in none of the nouns in (8a) is the constituent on the right a nominal (contra (7a)), and in none of the verbs in (8b) is the constituent on the left a verb (contra (7b)). Both patterns are productive. We thus see the freedom of projectivity in compounds. (8) a. [chang-duana]n long-short length b. [miann-shi]v face-test interview [dong-zuov]n move-make activity [yann-chan]v eye-envy envy [wai-yuv]n, outside-meet marital aair [shen-xing]v snake-walk walk like a snake

2.3. The disappearance of subcategorization It is generally recognized that if the selection or subcategorization of a verb has not yet been satised, it cannot be merged with another element. In (9a), for example, the verb mai3-le bought should be merged with the nominal shu book, to satisfy its subcategorization requirement, rather than the verb mai4 sell. However, in compounds like (9b), the root mai3 buy is indeed merged with mai4 sell, and the compound is acceptable. In (10a), the transitive verb like element zha fry is not merged with any internal argument. Instead, it is merged with the material-denoting element you oil. Similarly, in (10b), although the transitive verb-like element kan look at is merged with niao bird, the whole compound does not mean to look at a bird. In other words, niao here does not function like an internal argument to satisfy the subcategorization of kan. Instead, the compound means to look over like a bird does. The real internal argument is zhengge chengshi the whole city. (9) a. Ta mai3-le {shu/*mai4}. he buy-prf book/sell He bought books. b. yi zhuang mai3-mai4 one cl buy-sell a transaction of trade
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 175 (10) a. Wo yao you-zha zhexie jiaozi I will oil-fry these dumpling I will fry these dumplings with oil b. Ni keyi niao-kan zhengge chengshi. you can bird-look whole city You can look at the whole city like a bird does. These examples show that root merger does not care about subcategorization at all.

2.4. The issue of Case and theta role assignment Another well-known constraint in syntax is that a formal feature such as Case and theta roles cannot be licensed twice. For instance, if a transitive verb has been merged with a nominal as its internal argument, the resultant complex cannot take another nominal as an internal argument. This is shown in data like (11a). The formal features of the transitive verb xie write are satised by the object zi character and thus the nominal yi pian wenzhang an article cannot occur as another object. However, as pointed out by Chao (1968: 132), chu produce in (11b) takes ban version as its object, but the whole chu-ban publish behaves like a transitive verb, taking another nominal yi bu shu one cl book as an object. (11c) shows the same point. (11) a. *xie-le zi [dp yi write-prf character one b. chu-ban [dp yi bu produce-version one cl to publish a book c. fu-ze [dp zhe jian carry-duty this cl to take charge of this matter pian cl shu] book shi] matter wenzhang] article

Likewise, in (12), yan eye and hong red seem to have a subject-predicate relation, the former functioning as the unique argument of the latter. But the whole yan-hong envy takes bie-ren de caifu other peoples wealth as its internal argument. Since there is no more unlicensed formal features in the compound, it should not take another nominal as its internal argument, contrary to the fact. (12) Ta yan-hong [dp bie-ren de caifu] he eye-red other-people mod wealth He envies others wealth In an NV-like or VN-like compound, if the N-like part has a certain semantic relation with the complement of the compound, such as a
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

176 Niina Ning Zhang possessive or modication relation, one might claim that the former is raised from a complex nominal that also contains the latter. For instance, in (13a), wo I is the complement of the compound gen-zong followtrace, and it is also the possessor of the second part of this compound, zong trace. One might assume that zong is base-generated in the same complex with wo and then adjoins to gen follow, as illustrated in (13b). Such constructions are similar to the so-called classifying incorporation constructions discussed in Mithun (1984). (13) a. Ta zai gen-zong wo. he prog follow-trace I He is following me. b. Ta zai gen-zongi [wo ti]. However, in many compounds, no root has any semantic relation with any nominal external to the compound. For instance, in (11c), ze duty has no semantic relation with zhe jian shi this matter. Similarly, in (12), yan eye has no semantic relation with bie-ren de caifu others wealth. Such data are dierent from incorporation constructions, and thus the movement derivation hypothesis illustrated in (13b) does not apply to them.3 We can see in the above four subsections that certain syntactic constraints are violated in root merger. The following two subsections show that root merger is constrained in some ways that are not found in regular syntactic computation.

2.5 The eect of Lexical Integrity in movement One special constraint on root merger is that no element can be moved out of a compound. This has been captured by the well-accepted Lexical Integrity condition (Di Sciullo & Williams 1987, and Spencer 1991). The condition states that no syntactic process is allowed to refer to parts of a word. (14) a. Tamen yixiang fu-ze. they always carry-duty They are always responsible b. *Tamen yixiang lian ze dou fu. they always even duty also carry Intended: They are always even responsible. In (14a), fu-ze carry-duty ) responsible is a compound. (14b) shows that part of this compound, ze duty, cannot be preposed in the lian dou
3 Swedish compounds similar to (11b, c) and (12) are discussed in Josefsson (1998: sec. 4.2.1 and 4.2.2). She concludes that such data are derived by free root merger, rather than any syntactic Move (p. 77). One will see later in section 3 that we reach the same conclusion.

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 177 even also focus construction. If all words are derived in syntax, this special constraint on the movement of roots needs an account.

2.6 The eect of Lexical Integrity in pronominalization Another eect of the Lexical Integrity condition is that it is impossible to refer to part of a word by using an anaphoric device such as a pronoun. In (15), ta it cannot take cha tea, which is part of the compound cha-hu tea-pot, as its antecedent.
(15) *Ta xian na-le yi ba chai-hu, ranhou ba tai dao-ru beizi-li. he rst take-prf one cl tea-pot then ba it pour-in cup-in Intended: He rst took a tea-pot, and then poured the tea into a cup.

If roots are syntactic constituents, their inability to be antecedents of pronouns is puzzling. Summarizing, from the perspective of syntactic computation, we have seen six aspects of abnormality of root merger in Chinese. If there is a lexicon component in the computation system, and if all compounds are derived in the lexicon rather than syntax, we can treat all of these abnormalities as properties of lexicon-internal operations, using the assumed lexicon as a place to store the cases that we fail to explain. However, a syntactic account and a unied one is available for the abnormalities.

3. Root merger in the absence of categorial features I claim that all of the abnormal aspects of root merger can by accounted for by the hypothesis that roots do not have any syntactic features. Let us rst consider the projection abnormality presented in 2.1. In syntactic merger, projection means that the category of a complex is identical to one of its daughters. If roots do not have categorial features, they have nothing to project when they are merged together. The category of the resultant element is decided by the selecting functional element. Roots are categorized by combining with category-dening functional heads (Marantz 1997, 1999, Embick & Noyer 2005: 5, Borer 2005a: 21). I illustrate this dependency in (16) (f means functional

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

178 Niina Ning Zhang morpheme). The complex root is a nominal in (16a), an adjective in (16b), and a verb in (16c). Thus, strictly speaking, it is not that two adjectival roots form a noun compound in data like (1), since roots have no category. Instead, one should describe the compounding operation as the merger of two property-denoting roots. Second, it is a standard assumption that when two elements are merged, it is necessary to determine which of them projects, i.e. which one determines the syntactic category of the resulting structure (Chomsky 1994). How do we decide the projecting element? In set-merge of elements that have categorial features, the element that selects its sister, or the element whose uninterpretable features are valued by its sister, projects. For instance, in the phrase buy a book, the selection feature of the verb buy is satised by the DP a book, the V feature of buy projects, so that the whole phrase buy a book is a VP, rather than DP. As stated in Chomsky (2000), the label of the selector projects (p. 134), and computation is driven by a probe/selector of a label, which projects (p. 135). Since roots have no syntactic features, they do not have any uninterpretable features, and thus there is no issue of projection when one root is merged with another root. Thus the resultant complex can be any category, depending on the selecting functional category. This accounts for data like (5b) in 2.2. In pair-merge of elements that have categorial features, i.e., when a modier and the modied element are merged, it is the modied element that projects. Uriagereka (2006) proposes that a regular Merge operation produces a category with a new category label, but the Merge operation for modication is Merge without category labeling, so that the integration of a modier does not change the category of the resultant complex. In other words, before a modier is merged with the modied element, both already have categorial features. If roots do not have any categorial features, and if one root semantically modies the other, it makes no sense to talk about how the modication may keep the original categorial features. Thus, again, the resultant complex can be any category, depending on the selecting functional category. This accounts for data like (6b) in 2.2. Third, the selectional requirement or subcategorization of an element comes from the syntactic features of the element. In the absence of any such syntactic features, a root is able to merge with another root freely, if any semantic relation can be established between them in the language. This accounts for the abnormality of root merger discussed in 2.3. Fourth, the licensing of formal features, such as the assignment of Case and theta roles, is also a function of syntactic features, which roots do not have. In the absence of any such syntactic features, the merger of one root with another root is neither driven by nor leads to any licensing or
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 179 double licensing of Case or theta roles. This covers the fact discussed in 2.4. Fifth, elements that have no categorial features cannot move. The theoretical background of this claim is the following. Any element that undergoes overt movement must have its categorial features. In Chomsky (1995), all overt movement chains are composed of two sub-threads: a thread of formal features that take part in the required checking (CH <FF>) and a thread of categorial features (CH <CAT>). The latter thread does not involve feature checking, but it is simply the carrier of the former thread. In Chomskys terms, overt formal feature movement is always pied-piped with a categorial chain. This implies that in the absence of categorial features, no overt movement is possible. If roots do not have categorial features, they cannot move. Alternatively, one may assume that all types of syntactic movement are triggered by checking of uninterpretable features. Such features are syntactic. Roots do not have them. Thus roots do not have uninterpretable features. As we know, in a movement operation, either the traveler or the target, or both, have uninterpretable features, and only elements that have syntactic features are able to move and check the uninterpretable features of the target. Since roots do not have syntactic features, they cannot move. This captures the eect of Lexical Integrity in movement discussed in 2.5. Sixth, all pronouns agree with their antecedents in /-features (person, number, and gender). Although some pronouns are underspecied with certain /-features, they must be specied and agree with their antecedent in at least one /-feature. For instance, the Chinese pronoun ta he/she/it has no gender feature, but it does have at least the third person feature. /-features are syntactic features, which roots do not have. If roots cannot provide the /-features that pronouns agree with, they cannot be antecedents of pronouns. This captures the fact discussed in 2.6. In this section, we have accounted for the six abnormalities of root merger by a single assumption: roots do not have syntactic features. 4. Breakable compounds as products of two ways of merger In Chinese, some expressions with the same phonological forms behave like compounds in one context and phrases in another. Such expressions are called breakable compounds (lihe ci) in grammar books (e.g. Zhang 1957). Typically, the string is composed of a transitive verb-like element and an object-like element. The whole string looks like a V(erb)-O(bject) string. One example is the string dan xin carry heart ) worry (Xue 2001: 61). The following examples show that dan xin should be considered to be a word. The example in (17a) shows that the apparent VO string takes
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

180 Niina Ning Zhang the nominal zhe jian shi this matter as its object. This is similar to the case of root merger discussed in 2.4. It is similar to the examples in (11b) and (11c). (17b) shows that movement of xin is forbidden. This is the eect of the Lexical Integrity in root merger, discussed in 2.5. Moreover, (17c) shows that it is not possible to answer the question with dan alone. This indicates that dan is not a word. All of these examples show that dan xin should be considered a compound word, and its components roots.
(17) a. Ta hen dan xin zhe jian shi he very carry heart this CL matter He is very worried about this matter. b. *Xin, wo yi-dian dou bu dan _ zhe jian heart, I one-bit all not carry this CL Intended: I dont worry about this matter. c. Question: Ta dan xin zhe jian he carry heart this CL Is he worried about this? Answer: {Dan xin/* Dan (a)} carry heart/ carry interjection

shi matter shi ma? Matter q

The examples in (18) are dierent. The string dan xin in (18a) does not have any object, indicating that xin could be the object of dan, i.e., the former is an NP and latter is a verb, and they form a VP. (18b) shows that it is possible to move xin in topicalization. Thus there is no eect of Lexical Integrity. (18c) shows that it is possible to answer the question with dan alone. Thus dan here is a full-edged word, rather than a part of a word. These facts show that both dan and xin have syntactic features and thus dan xin here should be considered a phrase.
(18) a. Ta dan xin. he carry heart He was worried. b. Xin, wo yi-dian dou bu dan. heart, I one-bit all not carry I am not worried at all. c. Question: Ta dan xin ma? he carry heart q Is he worried? Answer: {Dan xin/ Dan a} carry heart/ carry interjection

Working in a syntactic approach to word formation, Xue (2001: 76) states that all words are derived in syntax. He further proposes that breakable compounds are derived by two structures: (19a) is for
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 181

compound-like forms, as in (17); and (19b), which is a regular phrase structure, is for phrase-like forms, as in (18). In (19a), both daughters, V and N, have syntactic features such as categorial features. One problem of this structure is that it does not explain how the formal features of the daughters are licensed. For instance, the complex in (17a) selects the object zhe jian shi this matter. If the selection relation of the lower V has already been satised by the N, how can its projection, the upper V, take another object? As claimed by Packard (2000: 123), breakable compounds are construal of either word or phrase, depending on the context. In my analysis, the word-like forms of breakable compounds are derived by root merger, as represented in (16c), and the phase-like forms are derived in the VP phrase represented in (19b). As expected, not only VO-like strings, but also other kinds of strings such as Subject-Predicate-like strings (e.g. xin fan heart vexed) and parallel combination strings (e.g. shui jiao sleep sleep) behave like compounds in one context and phrases in another. Our analysis of the VO-like strings also applies to these types of data.
(20) a. Baoyu xin-fan-le haoji tian. Baoyu heart-vexed-prf several day Baoyu has been vexed for several days. b. Baoyu, xin (hen) fan. Baoyu heart very vexed Speaking of Baoyu, he was (very) vexed. (21) a. Baoyu yijing shui-jiao-le. Baoyu already sleep-sleep-prf Baoyu has already slept. b. Baoyu shui-(zhe) jiao ne. Baoyu sleep-prog sleep prt Baoyu is sleeping.

In (20), the semantic relation between xin heart and fan vexed is similar to that between a subject and a predicate. In (20a), the two elements form a compound, which is followed by the perfective aspect sux le. In (20b), however, the two elements may be separated by the degree word hen very.
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

182 Niina Ning Zhang In this sentence, xin functions as a syntactic subject, and fan together with hen is the syntactic predicate. In (21), both shui and jiao mean sleep. In (21a), the two elements form a compound, which is followed by the perfective aspect sux le. In (21b), however, the two elements may be separated by the progressive aspect marker zhe. In this sentence, jiao is the cognate object of the verb shui-le (cf. English dance a dance). The above discussion shows that without a syntactic context, a breakable compound is ambiguous between the two possible merger levels: the merger of two roots and the merger of words or phrases. Thus, syntactic contexts not only decide the category of a root, but also give us clues of the merger level of two elements. The existence of breakable compounds in Chinese can be accounted for by the fact that both the verb and the object nominal of a VP can be bare (i.e., without any inection marker) in this language. When two bare forms are adjacent, only the context can tell whether or not they are roots. In languages in which nouns and verbs are always signaled by their inection markers (e.g. Semitic and Romance languages), we see no counterpart of the Chinese breakable compounds. 5. Summary Working on the computation of roots in Chinese compounds, this paper has presented six major dierences between the merger of roots in compound formation and the merger of elements in phrase structure formation: the existence of exocentric structures, the freedom of projectivity, the disappearance of subcategorization, the double licensing of formal features, the eect of Lexical Integrity in movement, and such an eect in pronominalization. All of these dierences have been accounted for by the hypothesis that roots do not have any syntactic features. Moreover, this study also discussed a case of merger level underspecication: without a syntactic context, an isolated string of two elements can be ambiguous between the root-merger that eventually gives rise to a verb and the merger of a verb and its object that gives rise to a VP. References
Arad, M. 2003. Locality constraints on the interpretation of roots: the case of Hebrew denomominal verbs. Natural Language and Linguistic Theory 21: 737 778. Barner, D. & Bale, A. 2002. No nouns, no verbs; Psycholinguistic arguments in favor of lexical underspecication. Lingua 112: 771791. Borer, H. 2005a. In name only. Oxford University Press. Borer, H. 2005b. The normal course of events. Oxford University Press.

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

Root merger in Chinese compounds 183

Chao, Y.-R. 1968. Yuyan wenti [Issues of language]. Taiwan Shangwu Press, Taipei. Chomsky, N. 1994. Bare Phrase Structure. MIT Working Papers in Linguistics. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by step: Essays on minimalist syntax in honor of Howard Lasnik, ed. R. Martin, D. Michaels & J. Uriagereka, 89155. Cambridge, MA.: MIT Press. Di Sciullo, A. M. & Williams, E. 1987. On the Denition of Word. Cambridge, MA: MIT Press. Embick, D. & Noyer, R. 2005. Distributed Morphology and the syntax/ morphology interface. To appear in The Oxford Handbook of Linguistic Interface, ed. G. Ramchand & C. Reiss. Oxford University Press. bregas, A. 2005. Universals and grammatical categories: A Distributed Fa Morphology analysis of Spanish colour terms. Morphology and Linguistic Typology, eds. G. Booij, E. Guevara, A. Ralli, S. Sgroi & S. Scalise. On-line Proceedings of the Fourth Mediterranean Morphology Meeting (MMM4) Catania 2123 September 2003, University of Bologna, 2005. Halle, M. & Marantz, A. 1993. Distributed morphology and the pieces of inection. The view from Building 20, eds. K. Hale & S. Keyser. Cambridge, MA: MIT Press. 111176. Harley, H. 2005. One-replacement, unaccusativity, acategorial roots, and Bare Phrase Structure. Harvard Working Papers on Linguistics. Vol. 9, ed. S. Gorbachov & A. Nevins. Harley, H. & Noyer, R. 2003. Distributed Morphology. The second GLOT International State-of-the-article book: The latest in linguistics, eds. L. Cheng & R. Sybesma: 463496. Studies in Generative Grammar 61. Berlin: Mouton de Gruyter. Huang, S.-F. 1998. Chinese as a headless language in compounding morphology. New approaches to Chinese word formation, ed. J. L. Packard. 261283. Berlin: de Gruyter. Irwin, P. 2006. Non-inectional word formation in Distributed Morphology: Featuring synthetic compounds. University of New Hampshire. Ms. Josefsson, G. 1995. The notion of word class and the internal make-up of words. Working Papers in Scandinavian Syntax 56: 145. Josefsson, G. 1998. Minimal words in a minimal syntax. Amsterdam/Philadelphia: John Benjamins Publishing Company. Marantz, A. 1997. No escape from syntax: Dont try morphological analysis in the privacy of your own lexicon. U. Penn. Working Papers in Linguistics 4, 2: 201225. Marantz, A. 1999. Creating words above and below little v. MIT. Ms. Matushansky, O. 2006. Head movement in linguistic theory. Linguistic Inquiry 37: 69109. Mithun, M. 1984. The evolution of noun incorporation. Language 60: 847894. Mukai, M. 2004. Headedness of compound words in Minimalist framework. University of Durham. Ms. Myers, J. To appear. Generative morphology as psycholinguistics. To appear in The mental lexicon: Core perspectives, eds. G. Libben & G. Jarema. Elsevier. Packard, J. L. 2000. The morphology of chinese: A linguistic and cognitive approach. Cambridge University Press. Roeper, T., Snyder, W., & Hiramatsu, K. 2001. Learnability in a Minimalist framework: Root compounds, merger, and the syntax-morphology interface. The process of language acquisition, ed. I. Lasser. Frankfurt: Peter Lang. Spencer, A. 1991. Morphological theory. Oxford: Blackwell.
The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.

184 Niina Ning Zhang

Uriagereka, J. 2006. Adjunct types. Paper presented at Workshop on adjuncts. The 29th GLOW Colloquium, Barcelona, March 58, 2006. Xue, N.-W. 2001. Dening and automatically identifying words in Chinese. University of Delaware. PhD diss. Zhang, S.-K. 1957. Lu elun hanyu goucifa [On Chinese word formation]. Zhongguo Yuwen 1957. Vol. 6. Received July 31, 2006 Accepted February 13, 2007 Niina Ning Zhang Graduate Institute of Linguistics National Chung Cheng University Min-Hsiung Chia-Yi 621 TAIWAN [email protected]

The author 2007. Journal compilation The Editorial Board of Studia Linguistica 2007.