VII
VO IDIOMS:
VARIATION AND REPRESENTATION
This chapter discusses idiom chunks, specifically the VO collocation type, e.g., keep tabs
on and kick the bucket in English or pai1 ma3pi4 (pat horse-ass) 'to flatter' and chi1 dou4fu3 (eat
tofu) 'to flirt' in Chinese, including the so-called 'possessive object' construction, such as pai1 ta1-de
ma3pi4 (pat his horse-ass) 'to kiss his ass'. I will demonstrate with relevant facts and argue within
the lexical functional theory, especially within its lexical mapping theory, that the ambiguous (literal and idiomatic) readings of idiom chunks cannot be adequately accounted for within the c-structure, the f-c-structure, or the thematic structure. I propose a solution that integrates Lakoff's (1987) 'motivation' account of idioms and lexical specifications in LFG. While the focus is on idioms of the VO type, the discussions should apply to idioms of all types; likewise, while the data in the chapter are from Chinese and English, the discussions should apply to other languages as well.
This chapter is organized as follows. Section 1 gives a definition of idioms and discusses the variation among the syntactic constraints that idiom phrases impose on the idiomatic interpretations. Section 2 then discusses the c-structure solution implied in the treatment of VO compounds by Chao (1968) and Li and Thompson (1981). Section 3 is a critical review of the f-structure account that C. Huang (1986, 1990a) argues for. Bresnan's (1982b) 'classical' analysis of idiom chunks, which involves the thematic structure and non-thematic functions, is examined in section 4. In section 5, I will propose a solution that views idioms as lexicalized metaphors within the overall framework of Lakoff (1987) and LFG. Section 6 provides an interactionist interpretation for the variation in the syntactic and semantic behavior of idioms; section 7 concludes the chapter.
1. IDIOMS AND SYNTACTIC CONSTRAINTS
The expressions covered under the term 'idiom' in the literature are diverse, to say the least (e.g., Everaert, van der Linden, Schenk, and Schreuder 1995). An idiom is defined in the discussion here as a phrase with an intended meaning different from the literal meaning of the whole phrase, and perhaps more importantly, such an idiomatic interpretation is available only when this phrase appears in a limited range of syntactic environments. Idioms thus have two defining characteristics: one, its non-literal meaning, and two, its syntactic constraints. Both of the two are predictable only to a limited extent. For an example, I will use probably the most famous idiom in the literature, kick
the bucket. Whether the idiomatic meaning of 'to die' or 'to lose (one's) life' may be considered
compositional or not, this meaning most definitely cannot be predicated from the literal meanings of its parts. What is also well-known is that this idiomatic meaning is obtainable only in certain strictly limited syntactic environments, as the following examples amply demonstrate. (Note that = indicates that the literal reading is available, # the idiomatic reading.)
1a. He kicked the red bucket. (=) b. He kicked the buckets. (=) c. He kicked his bucket. (=) d. He kicked a bucket. (=)
e. He kicked buckets. (=) f. He kicked three buckets. (=) g. The bucket was kicked by him. (=) h. It was the bucket that he kicked. (=) i. He kicked the barrel. (=)
j. He kicked the political bucket. (#?) k. He kicked the fucking bucket. (=,#)
This idiom is in fact among the most restricted, allowing almost no syntactic variation. As shown above, kick the bucket allows only an expletive modifier on the noun (1k) and any other variation of the syntactic environment would make the idiomatic reading unattainable. The Chinese idiom qiao4 bian4zi (stick up braid), which shares the same meaning of 'to die' or 'to lose (one's) life', is also similarly restrictive. Nonetheless, it is also not difficult to imagine some of the non-idiomatic sentences above used in a real discourse to hint at the non-idiomatic meaning, perhaps jokingly or sarcastically. The most obvious one is 1j, which is of course a creative use of the idiom. When used in an appropriate discourse context it would no doubt convey the intended idiomatic meaning that this person was politically finished, especially given the fact that the literal reading is simply unfeasible. Nonetheless, both the speaker and the hearer would also know that this is an innovation, not part of the conventionalized form of the idiom.1 (It is of course possible that in time such creative uses may 'diffuse' into the conventionalized idiom. See section 6 for more discussion on this point.) The idiomatic reading, being more marked, is thus more prominent over the literal reading and often 'blocks' the literal reading (see Zeevat (1995) for more discussion on this point).
Besides syntactic restrictions, idioms may also impose functional or pragmatic restrictions. Consider the idiom hold your horses for example, it must be used as a direct or indirect command or request, never as an assertion (Kaplan 1995:89). Break a leg, on the other hand, must be used as a direct command only. Kick the bucket, however, unlike the verb die, cannot be used as a command. The idiom is the Pope Catholic is most restricted functionally in that the idiomatic interpretation of 'most certainly' does not obtain unless the phrase functions as a rhetorical question.
2a. Hold your horses! (=,#)
b. I told you to hold your horses. (=,#) c. You held your horses. (=)
3a. Break a leg! (=,#)
b. I told her to break a leg. (=) c. She broke a leg. (=)
4a. Kick the bucket! (=)
b. I told you to kick the bucket. (=) c. He kicked the bucket. (=,#) 5a. Is the Pope Catholic? (=,#) b. The Pope is Catholic. (=) c. Is the Pope Christian? (=)
Since idioms vary greatly in terms of their individual semantic structure, they most certainly do not share the same syntactic and pragmatic constraints. It has been claimed, however, that the syntactic constraints of an idiom is to a large extent determined by the semantic relationship among its parts (Wasow, Sag, and Nunberg 1983). The question is of course to what extent. Note
that even the two regular, non-idiomatic verbs eat and devour, as it is well-known, though semantically very similar, if not identical, have very different syntactic requirements.2
6a. Lee ate.
b. Lee ate the eggs.
7a.*Lee devoured.
b. Lee devoured the eggs.
Syntactic variation, it seems, may be far greater among idioms with a similar semantic structure. Let's see some examples. The two idioms, kick someone's ass and kiss someone's ass, I will demonstrate, have a similar semantic structure. Given the idiomatic readings of 'to punish someone' and 'to flatter someone' respectively, one may argue that ass here does not bear a semantic role. However, the following sentences, all allowing the idiomatic readings, would indicate that ass in both idioms has clearly identifiable, independent semantic content, referring to the person being punished or flattered in an abstract sense.
8a. I enjoyed kicking his white/black ass. (=,#) b. I enjoyed kicking his fat/bony ass. (=,#)
c. You mean you kicked the king's royal ass? (=,#) 9a. No way I would kiss his white/black ass. (=,#) b. No way I would kiss his fat/bony ass. (=,#) c. You mean you kissed the king's royal ass? (=,#)
The semantic content and the semantic relationships among the parts of the two idioms thus do seem to be parallel. However, there are differences in their syntactic and morpholexical behavior and the differences seem arbitrary.
10a. I'm sure you'll kick ass. (#) b.?I'm sure you'll kiss ass.
11a. Lee got/had his ass kicked. (=,#) b. Lee got/had his ass kissed. (=) 12a.?What an ass-kicking player! b. What an ass-kissing son of bitch! 13a. This is some kick-ass fast car! b.*What a kiss-ass lousy guy!
14a.*brown-footer/*brown-footing b. brown-noser, brown-nosing
Perhaps a more dramatic pair of examples is hold your horses and hold your breath. The semantic relationships among the parts are identical, but the former does not seem to allow negation, while the latter must be negated to have the idiomatic reading. Furthermore, while the former must be used as a command or request, the latter has no such functional restrictions.
b. Don't hold your horses. (=)
c. I told you not to hold your horses. (=) d. He's holding his horses. (=)
e. He's not holding his horses. (=) 16a. Hold your breath. (=)
b. Don't hold your breath. (=,#)
c. I told you not to hold your breath. (=,#) d. He's holding his breath. (=)
e. He's not holding his breath. (=,#)
I will now illustrate more extensively and systematically (but most certainly not exhaustively) the range of syntactic and functional restrictions that VO idioms may impose in Chinese. The two examples used here are chi1 dou4fu3 (eat tofu) 'to flirt (with)' or 'to take (sexual) advantage of' and peng4 ding1zi (knock-against nail) 'to be rejected' or 'to face a rejection'.
A. Syntactic behavior of O in VO idioms: I. Modification of O
a. numerical quantification (number + measure word, e.g., kuai4 'piece' for tofu or classifier
gen1 for nails.)
17. Ta1 chi1 le yi1 kuai4 dou4fu3. (=) He ate a piece of tofu.
18. Ta1 peng4 le yi1 gen1 ding1zi. (=,#) He suffered a rejection.
b. mass quantification, e.g., bu4shao3 'a lot of'. 19. Ni3 chi1 le ta1 bu4shao3 dou4fu3. (=,#) You flirted with her quite a bit.
20. Ni3 peng4 le ta1 bu4shao3 ding1zi. (=,#) You were rejected by her quite a few times.
c. adjective, e.g., nen4 'tender' and ruan3 'soft', without DE (note however, dou4fu3 can only be modified by nen4, ding1zi only be ruan3. Neither idiom takes a wide range of adjectives.)
21. Ta1 zhuan1 chi1 nen4 dou4fu3. (=,#) He flirts with the young ones only. 22. Ta1 peng4 le yi1 ge ruan3 ding1zi. (=,#) He was subtly rejected.
d. adjective with DE
23. Ta1 zhuan1 chi1 nen4 de dou4fu3. (=) He only eats tender tofu.
24. Ta1 peng4 le yi1 ge ruan3 de ding1zi. (=) He knocked against a soft nail.
e. zhe4/na4 zhong3 'this/that kind'
He won't flirt like this.
26. Ta1 bu4 hui4 peng4 zhe4 zhong3 ding1zi. (=,#) He won't face this kind of rejection.
f. determiner
27. Ta1 chi1 le zhe4 dou4fu3. (=) He ate this tofu.
28. Ta1 peng4 le zhe4 ding1zi. (=,#) He suffered this rejection.
g. time phrase, e.g., zuo2tian1 'yesterday', with DE
29. Zuo2tian1 de dou4fu3 hai2 mei2 chi1 gou4 ma1? (=,#) Didn't you do enough flirting yesterday?
30. Zuo2tian1 de ding1zi mei2 peng4 gou4 a? (=,#) Didn't you get enough rejections yesterday?
h. duration phrase, e.g., ban4tian1 'quite a while', with or without DE 31. Ta1 chi1 le ban4tian de dou4fu3. (=,#)
He flirted for quite a while.
32. Ta1 peng4 le ban4tian1 de ding1zi. (=,#) He suffered rejections for quite a while.
i. frequency phrase, e.g., san1 ci4 'three times', with or without DE 33. Ta1 chi1 le san1 ci4 dou4fu3. (=,#)
He flirted three times.
34. Ta1 peng4 le san1 ci4 ding1zi. (=,#) He was rejected three times.
j. possessive NP with DE
35. Ta1 chi1 wo3 de dou4fu3. (=,#) He flirted with me.
36. Ta1 peng4 le wo3 de ding1zi. (=,#) He was rejected by me.
k. possessive NP without DE 37. Ta1 chi1 wo3 dou4fu3. (=,#) He flirted with me.
38.*Ta1 peng4 wo3 ding1zi. II. 'Movement' of O
l. Bare topic
39. Dou4fu3 ta1 chi1 duo1 le. (=) Tofu, he ate quite a bit.
40. Ding1zi ta1 peng4 duo1 le. (=,#) He suffered a lot of rejections. m. Modified topic
41. Zhe4 zhong3 dou4fu3 ni3 ye3 gan3 chi1? (=,#) How dare you flirt like this?
42. Zhe4 zhong3 ding1zi ta1 chang2 peng4. (=,#) He often gets this kind of rejections.
n. BA construction
43. Ta1 ba3 dou4fu3 chi1 le. (=) He ate the tofu.
44. Ta1 ba3 ding1zi peng4 le. (=) He knocked against the nail. o. BEI construction
45. Ta1 de dou4fu3 bei4 ni3 chi1jin4 le. (=,#)
She has been fully taken advantage of by you (sexually). 46. Ta1 de ding1zi bei4 ni3 peng4jin4 le. (=)
Her nails were all knocked against by you. p. cleft
47. Ta1 chi1 de ke3 shi4 ni3 de dou4fu3. (=,#) It was you after all that he flirted with. 48. Ta1 peng4 de ke3 shi4 ni3 de ding1zi. (=) It was your nails that he touched.
q. relativized O
49. Ni3 chi1 de dou4fu3 hai2 bu2 gou4 duo1 a? (=,#) Haven't you done enough flirtations?
50. Ni3 peng4 de ding1zi hai2 bu2 gou4 duo1 a? (=,#) Haven't you had enough rejections?
III. Anaphora of O
r. verb copying (within a sentence):
51. Ni3 chi1 ta1 de dou4fu3 chi1 le ban4tian1. (=,#) You have been flirting with her for quite a while. 52. Ni3 peng4 ta1 de ding1zi peng4 le ban4tian1. (=) You have been touching her nails for quite a while. s. Purposive clause
53. Ta1 zhuan1 zhao3 nen4 dou4fu3 chi1. (=,#) He seeks out the young ones to flirt with. 54. Ta1 zhuan1 zhao3 ruan3 ding1zi peng4. (=) He seeks out the soft nails to touch.
t. Discourse recovery
55. Ta1 hen3 hui4 chi1 dou4fu3, chi1 de hen3 ji4qiao3. Bei4 chi1 de ren2 gen1ben3 bu4 zhi1dao4. (=,#)
He is very good at flirting, and does so skillfully. Sometimes, the person he's flirting with doesn't even realize it.
56. Wo3 kan4 ta1 lao3shi4 peng4 ding1zi. Wei4she2me ni3 hen3shao3 peng4? (=,#)
happens to you?
B. Syntactic Behavior of V in VO Idioms u. aspect marker
57. Ni3 chi1 guo4 ta1 de dou4fu3. (=,#) You have flirted with her before. 58. Ni3 peng4 guo4 ta1 de ding1zi. (=,#) You have been rejected by her before. v. resultative
59. Ni3 chi1-bu2-dao4 ta1 de dou4fu3. (=,#) You wouldn't get to flirt with her.
60. Ni3 peng4-bu2-dao4 ta1 de ding1zi. (=,#) You wouldn't get rejected by her.
w. reduplication
61. Wo3 chi1 le chi1 ta1 de dou4fu3. (=,#) I flirted with her a bit.
62. Wo3 peng4 le peng4 ta1 de ding1zi. (=) I touched her nails a bit.
x. prefix hao3 (tough construction)
63. Ta1 de dou4fu3 ke3 bu4 hao3chi1. (=,#) It's tough to flirt with her.
64. Ta1 de ding1zi ke3 bu4 hao3peng4. (=) Her nails are tough to touch.
C. Discourse Functions y. Request/command
65. Chi1 ta1 de dou4fu3! (=,#) Flirt with her!
66. Peng4 ta1 de ding1zi! (=) Knock against her nails. z. Question
67. Ni3 chi1 le ta1 de dou4fu3 ma? (=,#) Did you flirt with her?
68. Ni3 peng4 le ta1 de ding1zi ma? (=,#) Did you get rejected by her?
With this systematic review of even just two idioms, it should be clear enough that the syntactic constraints on the idiomatic interpretations are first of all real, and secondly hard to predict from either the meaning of the idiom as a whole or the semantic relationships among the parts. An adequate treatment of idiom phrases therefore must account for not only the relationship between the idiomatic meaning and the literal parts but also the allowable syntactic environments in which the idiomatic reading may obtain.
2. THE C-STRUCTURE SOLUTION
There are two fundamentally different approaches for the analysis of idioms: one approach considers idioms similar to words as basic lexical units, and the other treats idioms as phrases whose internal syntactic structures participate in the structure of the sentence (e.g., Stock 1987). A c-structure solution, where the ambiguity between a literal reading and an idiomatic reading is accounted for by assigning two distinct tree structures to the two readings, is certainly of the first approach. As far as I know, such a solution has never been seriously or explicitly proposed within LFG. However, studies of VO compounds that treat VO idioms as compounds in essence imply that the idiomatic reading, where the idiom is a compound and thus of the lexical category V, has a tree structure distinct from that of its literal reading, where the same VO string forms a phrasal category, VP, as shown in 69-70.
69. Ta1 xi3huan1 [Vchi-dou4fu3V]
he like flirt He likes to flirt.
70. Ta1 xi3huan1 [VP[Vchi1V][NPdo4fu3NP]VP]
he like eat tofu He likes to eat tofu.
First of all, if the lexical integrity hypothesis is assumed as in all earlier chapters, then the status of any given VO sequence is easily distinguished: a string that allows no syntactic rule to affect its sub-parts is, by definition, a word, and thus not a phrase, idiomatic or not. There are indeed many such compounds, e.g., guan1xin1 (close-heart) 'to be concerned about' and na2shou3 (take hand) 'to be good at'. However, as demonstrated in Chapter 3, there are also plenty of genuine idiom phrases, such as chi1 dou4fu3. Thus, under the lexical integrity hypothesis, the ambiguity in idiom phrases cannot be accounted for by c-structure variations, for the two readings are assigned the identical phrasal category VP.
71. Ta1 xi3huan1 [VP[Vchi1V][NPnen4 dou4fu3NP]VP]
he like eat tender tofu He likes to eat tender tofu. OR He likes to flirt with the young ones.
On the other hand, if we follow Chao's (1968) proposal that idiomaticity of a VO sequence is sufficient for its compound status and thus abandon lexical integrity, then indeed chi1 dou4fu3 would pose the two distinct c-structures 69 and 70. In Chao (1968) and Li and Thompson (1981) many idiomatic VO phrases are indeed taken to be VO compounds instead of phrases. I have discussed the drawbacks of this confusion between lexicon and syntax in Chapter 3. Here I will simply demonstrate why such a position is unworkable. This account is immediately in trouble in the presence of the following examples.
72. Ta1 xi3huan1 chi1 ni3/ta1/Ma3li4/na4 ge ren2-de dou4fu3. He likes to flirt with you/her/Mary/that person.
The possessive NPs that may intervene the subparts of the idiom are productive. Furthermore, as I have demonstrated in the previous section, the possible intervening elements are certainly not limited to possessive NPs. Since the lexicon of a grammar must always be limited and
constrained, this account, which renders an infinitely large lexicon, is impossible.
3. THE F-STRUCTURE SOLUTION
C. Huang's (1986) dissertation contains an f-structure solution for the ambiguity of idiom chunks. Within the formal model of LFG, given that grammatical functions play a central role in grammatical description and that the idiom shares an identical c-structure with the literal reading (see section 1 and the discussion in section 2.A of C. Huang 1990a:265-267), the next logical step is certainly to look for a solution in the f-structure. This section is a critical review of the f-structure account further revised and formalized in C. Huang (1990a), where the two readings of an idiom chunk, for example 73a-b, are assigned two distinct f-structures.
73. San1bai2 chi1 cu4. Sanbai eat vinegar a. Sanbai eats vinegar. b. Sanbai gets jealous.
First of all, I should point out the inconsistency between the conclusion C. Huang makes from his observation and the actual formulation of his analysis. He observes that the idiomatic reading and the literal reading are radically different in their semantics, but concludes that 'there is
no evidence of any syntactic distinction between them from which the semantic differences can be
derived (C. Huang 1990a:263, emphasis added). What he proposes, nevertheless, is a syntax-based solution, or in his words 'a lexically-based LFG analysis in which the differences between the two constructions are accounted for in terms of differences in f-structure' (C. Huang 1990a:263, emphasis added). The f-structure, however, is part of the syntax proper. One of the most important aspects of f-structure autonomy is that grammatical functions cannot be semantically derived. The relation between thematic structure (or the lexical semantic structure) and the functional structure is that of mapping or correspondence, not derivation. The following is LFG's model of semantics-syntax interface, in contrast with that of the transformational model, from Bresnan (1995).
74a. The LFG Model of Syntactic Projection
lexical semantics
Lexico-semantic projection a-structure
Lexico-syntactic projection
syntactic structure
b. The Transformational Model of Syntactic Projection
lexical semantics
Lexicon a-structure
Syntactic projection initial syntactic structure
Syntactic transformations final syntactic structure
In LFG the interface a-structure maps to the syntactic structure, i.e., f-structure, as shown in 74. The c-structure and f-structure are the two parallel planes of syntax, and the f-structure is in fact the core of syntax. Huang's solution is thus well within the domain of syntax. Consequently, if his conclusion that no syntactic distinction exists between the idiom and the literal reading is correct, then his analysis, which assigns different f-structures to the two readings, can be expected to have difficulties. Let's take a closer look at this f-structure account: 75a-b are assigned the following two distinct f-structures.
75. San1bai2 chi1 cu4. a. Sanbai eats vinegar.
a-f
[ SUBJ [ PRED 'SANBAI' ] PRED 'EAT <SUBJ OBJ>' OBJ [ PRED 'VINEGAR' ] ]
b. Sanbai is jealous.
b-f
[ SUBJ [ PRED 'SANBAI' ] VMORF CHI PRED 'BE-JEALOUS <SUBJ>' ]
Note that the idiom chi1 cu4 (eat vinegar) 'to be jealous' in 75b-f does not have the function OBJ, which the literal cu4 'vinegar' necessarily encodes, as in 75a-f. Idiomatic cu4 is not recognized as an argument in 75b-f at all. This position is based on three syntactic tests: coordination, wh-question formation, and topicalization (C. Huang 1990a:269-270). Let's go over them one by one.
76. Li3si4 chi1 pang2xie4 gen1 cu4. Lee eat crab and vinegar a. Lee eats crabs and vinegar. b.*Lee eats crabs and is jealous.
Coordination involves parallel constructions sharing an identical grammatical function, for example, OBJ pang2xie4 and OBJ cu4 in 76a. Since pang2xie4, an argument, cannot be conjoined with the idiomatic cu4 in 76b, Huang concludes that cu4 must not be argument in the idiom chi1
cu4. However, coordination requires more than parallel grammatical functions; 76b thus may be
ruled out due to another violation.
77. Ta1 kai1 le men2 gen1 tai2deng1. he open/turn-on PERF door and lamp a. He opened the door and the lamp.
b.*He opened the door and turned on the lamp. 78.*I admire Mary and honesty.
Clearly, tai2deng1 is an OBJ argument; 77b is ruled out because kai1 is allowed the reading
of 'to open' due to the adjacent 'door'. Similarly, 78 is ill-formed because 'Mary' and 'honesty' are not semantically compatible as conjoined elements. Therefore, it is entirely plausible for the
idiomatic cu4 to be a referential argument OBJ and that 77b is ruled out due to the semantic incompatibility between pang2xie4 'crab' and the idiomatic cu4 and/or the ambiguous readings of
chi1. An equally plausible, and compatible, explanation is due to the defining characteristic of
idioms that the idiomatic reading is obtainable only in restricted syntactic environments. Thus, 77b is ill-formed simply because idiomatic cu4 allows no conjunction, period. Only this last explanation of syntactic constraint accounts for the following sentence's non-idiomatic reading. Given the equal idiomatic, non-argument status of dou4fu3 and cu4 in Huang's account, the idiomatic reading of 79b should be allowed as well, but is not.
79. Li3si4 chi1 dou4fu3 gen1 cu4. Lee eat tofu and vinegar a. Lee eats tofu and vinegar. b.*Lee flirts and gets jealous.
The next test Huang employs is wh-questions. It is a fact, as C. Huang (1990a:270) points out, that syntactic arguments can form wh-questions. He thus concludes, from the unavailability of 80b below, that qi4 in the idiom sheng1 qi4 (generate air) 'to be angry' must not be a syntactic argument.
80. Ta1 sheng1 she2me? she generate what
a. What does she generate? b.*What is she angry with?
c.*What does she give birth to a baby? 81. Ta1 sheng1 xiao3hai2.
she generate baby She gives birth to a baby.
The problem with this particular argument is with its logic. Given the necessary condition that sheng1 and qi4 must co-occur in certain restricted syntactic environments for the idiomatic reading to obtain, 80 of course cannot possibly maintain the idiomatic reading with the position of
qi4 replaced by a wh-word. By the same token, non-idiomatic xiao3hai2 in 81 can be replaced by a
wh-word, as in 80; nonetheless, 80 certainly cannot have the same meaning as 81, shown in 80c. The last test is topicalization. It seems that the NP of an idiom cannot be topicalized and thus does not behave like an argument.
82. Cu4, ta1 chi1. vinegar he eat a. Vinegar, he eats. b.*He is jealous.
What Huang has overlooked is that although the bare N, like cu4, in VO idioms indeed usually does not topicalize, often when it is modified in some way, topicalization may indeed take place, indicating the referentiality and argumenthood of the head noun.
83. Zhei4 zhong3 cu4 ni3 ye3 chi1, tai4 hai3ziqi4 le. this kind vinegar you also eat too childish PTCL You are jealous because of this, that's childish.
84. Zhei4 zhong3 qi4 bu4 zhi2de2 sheng1. this kind air not worth generate It's not worth it to be angry about this. 85. Ta1 de dou4fu3 ni3 ye3 gan3 chi1? she POSS tofu you also dare eat You dare take sexual advantage of her?
Treating the idiomatic NP as a non-argument, Huang's analysis also fails to account for passivizable idioms, for example the idiom chu1 yang2xiang4 that Huang (1990a:282) specifically mentions in his analysis and similarly the idiom diu1 lian3 'to lose face'.
86a. Ta1 chu1jin4 le yang2xiang4. he produce-all PERF foreign-picture He totally made an ass out of himself. b. Yang2xiang4 bei4 ta1 chu1jin4 le. foreign-picture by he produce-finish PTCL What an ass he totally made out of himself. 87a. Ta1 shu1 le qian2 hai2 diu1 le lian3. he lose PERF money and lose PERF face He lost money and lost face.
b. Lian3 bei4 ta1 diu1-guang1 le. face by he lose-empty PTCL All honor is lost by him.
Within the pre-LMT model of LFG, which Huang seems to assume (see Huang's (1990a:272) discussion on passive), the passive lexical rule (88) converts OBJ to SUBJ. Therefore, Huang's analysis predicts, incorrectly, that VO idioms are non-passivizable. Since yang2xiang4 and
lian3 bear no grammatical function OBJ, they cannot possibly be converted to passive SUBJ; see
88.
88. Passive (Bresnan 1982b): SUBJ /OBL OBJ SUBJ
89. Passive (Bresnan 1989): < ..>
Within the current theory of lexical mapping (see Chapter V for details), yang2xiang4 and
lian3, treated as non-arguments in Huang's analysis, likewise cannot be mapped to SUBJ as the
passive morpholexical rule (89) suppresses the highest role. Passive 86b and 87b are thus still predicted to be non-existent. Following this line of argument, all lexical processes that involve either the function OBJ in the pre-LMT model or a thematic role linked to this OBJ in LMT are ruled out in Huang's analysis.
Another distinctive feature of Huang's analysis is that the noun, rather than the verb, of the idiom is considered the lexical head. This is accomplished by stipulations of additional c-structure rules and lexical entries for both the noun and the verb.
90a. chi12: V, VMORF = CHI
VMORF =c CHI
91. VP V NP =
It is rather common to pose homophone entries for idiomatic elements, although it does lead to a proliferation of lexical entries.3 Another problem is with the c-structure rule (91), which is needed for the sole purpose of generating the stipulated f-structure of VO idioms like chi1 cu4. Note that this rule duplicates the c-structure constructed by the regular VP rule (92), the only difference being the schemata specifying the NP head.
92. VP V NP = OBJ=
Consider this question: which of the two rules should actually apply to a given applicable string? In the case of idiom phrases, both rules need to apply to produce the ambiguous readings, but for phrases with a literal reading only, the idiom VP rule must then be barred. Consequently, rule marking must be introduced into the LFG formalism, an undesirable feature for any grammatical framework with an X-bar scheme.
Another undesirable consequence of this analysis is that any adjunctive element of the noun necessarily has the entire idiom as its scope, not just the noun. Take chi1 dou4fu3 for example. The modifier nen4 'tender' of tofu in 93 below would have the entire idiom as its scope, as shown in 93-f. This is incorrect for the sentence does not mean 'Lee only flirts tenderly' or 'Lee only flirts youthfully' as the scope of nen4 is clearly limited to the noun, tofu, that it modifies.
93. Li3si4 zhuan1 chi1 nen4 dou4fu3. Lee only eat tender tofu Lee only flirts with the young ones.
-f
SUBJ [ PRED 'LEE' ] ADJ {[ PRED 'ONLY' ] [ PRED 'TENDER' ]} VMORF CHI PRED 'FLIRT <SUBJ>'
The fact that the noun in the idiom may be independently modified also argues for its referentiality, and thus against Huang's treatment of it as a non-argument. This is confirmed by the fact that the noun in the idiom may be referred to by an empty anaphora.
94. Ta1 ai4 chi1 dou4fu3, hai2 zhuan1 chi1 nen4 de e. he like eat tofu and only eat tender COMP He likes to flirt and only with the young ones. 95. Ta1 ai4 pai1 ma3pi4, dan4 pai1 e de bu4 hao3. he like pat horse-ass but pat COMP not good He likes to kiss ass, but he's not good at it.
Huang's analysis is further complicated by his treatment of the so-called 'possessive objects' (POBJ). The possessive Mary in 96a has the function POSS within the function OBJ; however, in the idiom (96b), it is treated as an OBJgoal independently. Additional NP rule (97a) is thus needed,
other than the regular NP rule (97b) that assigns POSS to NP-de, to build an OBL function out of the NP-de phrase.
96. Li3si4 chi1 Ma3li4-de dou4fu3. Lee eat Mary-POSS tofu a. Lee eats Mary's tofu.
a-f SUBJ [ PRED 'LEE' ] PRED 'EAT <SUBJ OBJ>'
OBJ POSS [ 'MARY' ] PRED 'TOFU'
b. Lee flirts with Mary.
b-f SUBJ [ PRED 'LEE' ] VMORF CHI PRED 'FLIRT <SUBJ OBL>' OBL [ PRED 'MARY' ]
97a. NP (NP) (XP*) (CL) N POSS= ADJ = b. NP (NP) (XP*) (CL) N OBL= ADJ =
The use of the term 'possessive object' to refer to NP1 of [NP1 de NP2] within a VO idiom finds its origin in Chao's (1968:321) discussions on VO compounds, where the term refers to 'an apparent possessive modifier, instead of the object, (that) represents the goal'. It is rather curious for Huang to retain this term, since he sees [NP1 de] as neither possessive nor objective. Rather he identifies it as an oblique grammatical function, or OBL, which bears the goal role. The only piece of evidence offered for this function assignment is from such 'synonymous' pairs (Huang 1990a:-274):
98a. Wo3 jian4 le ta1 de mian4. I see PERF she POSS face I met her.
b. Wo3 gen1 ta1 jian4 le mian4. I with she see PERF face I met her.
99a. Yung2niang3 sheng1 San1bai3 de qi4. Yunniang generate Sanbai POSS air Yunniang is angry with Sanbai.
b. Yung2niang3 gen1 San1bai3 sheng1 qi4. Yunniang with Sanbai generate air Yunniang is angry with Sanbai.
Such pairs are reminiscent of the similar observation made in Chao (1968:321-322, 430-431). However, Chao (1968:321-2) has also noted that many of these idioms do not paraphrase into a corresponding sentence where NP1 occurs in a PP, goal-like or not. For example, cai1 tai2 (dismantle platform) 'to spoil someone's plan', ge2 ming4 (revolt life) 'to revolt (against someone)', as well as some of the a dozen or so idioms listed in Huang's account like chi1 dou4fu3 (eat tofu) 'to take (sexual) advantage of' and chi1 bing1qi2lin2 (eat ice cream) 'to take (visual) advantage of'. Worse still, again as noted by Chao (1968:322), often the corresponding sentences with NP1 in a PP would have entirely different meaning, as shown in 100-101 below.
100. Ta1 gao4 Ma3li4 de zhuang4. he tell Mary POSS suit
He made a complaint against Mary.
101. Ta1 gen1/xiang4/dui4 Ma3li4 gao4 zhuang4. He with/towards/to Mary tell suit
He made a complaint to Mary (again someone else).
The fundamental problem is that it is altogether unreliable to determine the grammatical function of an element in a particular construction by its paraphrase in another construction. The active-passive paraphrase, dative shift, locative inversion, and cleft, to name just a few, are obvious examples of paraphrase constructions that assign different grammatical functions to the same semantic role. Some example sentences are shown in 102-104 below.
102a. John loaded grapes in the truck. OBJ OBL
b. Grapes were loaded in the truck. SUBJ OBL
c. John loaded the truck with grapes. OBJ OBL d. The truck was loaded with grapes. SUBJ OBL
103a. My wife Koto is sitting among the CEO's. SUBJ OBL
b. Among the CEO's is sitting my wife Koto. SUBJ OBJ
104a. I handed the award to her. OBJ OBL b. I handed her the award. OBJ OBJ
An even more basic question is whether the possessive NP in a VO idiom behaves like an argument or like other regular possessor NPs. Let's look at the various tests Huang employs for its argument status. The first two are wh-question formation and reflexive pronoun (C. Huang 1990a:270-271).
105. Ta1 sheng1 shei2-de qi4? he generate who POSS air Who is he angry with?
106. San3bai2 sheng1 zi4ji3-de qi4. Sanbai generate self POSS air Sanbai is angry with himself.
Huang (1990a:271) argues that the possessive NP in idioms can be questioned and replaced by a reflexive pronoun suggests that NP1 is a referential argument. Referential, yes; argument, not necessarily. Look at the following two sentences.
107. Ta1 xiang3 shei2-de shi4? he think who POSS affair Whose affairs does he think about? 108. Ta1 xiang3 zi4ji3-de shi4. he think self POSS affair He thinks about his own affairs.
The possessive NP here passes the same tests, but certainly it cannot be considered an argument of xiang3 'to think'. Huang further employs evidence from selectional restrictions to argue for NP1's argumenthood. His argument goes likes this: '..the predicate jian-mian 'to meet' selects a human object...The fact that the idiom chunk jian-mian 'to meet' imposes selectional restrictions on NP1 indicates that NP1 is an argument of the idiom chunk, regardless of its position in a syntactic tree' (C. Huang 1990a:272).
109. Ta1 jian4 le zuo1zi de mian4. he see PERF table POSS face/surface a. He saw the surface of the table.
b.*He met the table.
Huang's argument presupposes that jian4 mian4 as an idiom is semantically a single non-decomposable predicate that requires a human object. Hence, his conclusion that NP1 is an argument (and NP2 mian4 is not) is already in the presupposition. Nonetheless, the anomaly of 109b is not necessarily due to NP1 zuo1zi's violation of selection restriction of the idiom. Note that
mian4 is itself ambiguous between 'surface' and 'face', the latter being that of a human. Thus, the
two compounds mian4zhi3 (face-paper) 'facial tissue-paper' and zuo1mian4 'table-top' involve the two different readings of mian4. The problem with 109b is therefore the anomalous reading 'table's (human) face'. However, to be fair, this kind of explanation is not available for VO idioms where the object noun like dou4fu3 'tofu' is not ambiguous.
110. Ta1 ai4 chi1 na4 jia1 dian4 de dou4fu3. he like eat that CLS shop POSS tofu a. He likes to eat the tofu from that store. b.*He likes to flirt with that store.
We therefore still have to go back to the syntactically and semantically restricted nature of idioms for a general explanation. As mentioned earlier, one defining character of idioms is that the idiomatic reading is obtainable only in restricted syntactic environments. One of the restrictions for VO idioms like chi1 dou4fu3 'to flirt (with)' and jian4 mian4 'to meet' is that the noun can only take human possessors. (Note that some VO idioms do not allow possessors at all, for example qiao4
bian4zi and kick the bucket). This restriction is also not unreasonable in itself, given that nouns do
often allow only a restricted, and sometimes arbitrary, range of possessors. See the following two sets of examples.
111a. the man's face b. the dog's face c.?the fish's face d.*the book's face e.*the mountain's face f.?IBM's face
112a. the man's words b.?the dog's words c.?the fish's words d. the book's words e.*the mountain's words f. IBM's words
Likewise, the fact that chi1 dou4fu3 as an idiom also selects a very small number of adjectives on dou4fu3, for example nen4 'tender' but not ruan3 'soft', cannot be taken to indicate that these adjectives as thus arguments. Next Huang uses the test of conjunction to argue that NP1 in the idiom [V NP1 de NP2] is not a case of possessor. In the example of 113, where if Wangwu, obviously the possessor of pang2xie4 'crab', is also the possessor of the idiomatic cu4 'vinegar', the conjoined phrase should still allow the idiomatic reading. The fact that 114b is not available is thus taken to indicate that Wangwu does not have the same possessor function to idiomatic cu4.
113. Li3si4 chi1 Wang2wu3 de pang2xie4 gen1 cu4. Lee eat Wangwu POSS crab and vinegar
a. Lee eats Wangwu's crabs and vinegar.
b.*Lee eats Wangwu's crabs and is jealous of him.
Again, two functions being identical is merely one of the necessary conditions for a well-formed conjunction, not the only one. As mentioned earlier, there are two compatible explanations here: 1) pang2xie4 'crab' and idiomatic cu4 are semantically incompatible for conjunction, and 2) the idiom does not allow cu4 to be conjoined, period. This is merely one of the many constraints that this idiom imposes. Thus, that 113b is not available is simply because the idiomatic reading of
chi1 cu4 cannot obtain in a syntactic environment where cu4 is conjoined. Wangwu remains the
possessor of cu4, idiomatic or not. This explains why 114b below is not available, while Huang's account, where Wangwu as OBL in both idioms should allow conjunction, would predict that it is well-formed.
114. Li3si4 chi1 Wang2wu3 de do4fu3 gen1 cu4. Lee eat Wangwu POSS tofu and vinegar a. Lee eats Wangwu's tofu and vinegar.
b.*Lee flirts with Wangwu and is jealous of him.
The more convincing argument that Huang provides against treating NP1 as a regular possessor comes from anaphora. He points out the difference in the binding relations between regular phrases and the idiom phrase. Pay close attention to the binding relations in the following sentences.
115a. San1bai2i xi3huan1 zi4ji3i-de shi1.
Sanbai like self POSS poem Sanbai likes his own poem.
b. San1bai2i xi3huan1 ta1i/j-de shi1.
Sanbai like s/he POSS poem Sanbai likes his poem.
116a. San1bai2i chi1 zi4ji3i-de cu4.
Sanbai eat self POSS vinegar Sanbai is jealous with himself. b. San1bai2i chi1 ta1*i/j-de cu4.
Sanbai eat s/he POSS vinegar Sanbai is jealous with him.
Reflexive zi4ji3 as NP1 always refers back to the matrix subject whether the verb phrase is
idiomatic or literal, as shown in 115a and 116a. No drama here. However, while a non-reflexive pronoun as NP1 is free within the regular sentence (115b), it must not be bound with the matrix subject in the sentence of a VO idiom (116b). The two different f-structures in Huang's analysis for the regular possessor and the idiomatic possessor, which is an OBL argument, provides an explanation.
117. San1bai2i chi1 ta1i/j-de cu4. (literal reading)
Sanbai eats his vinegar.
-f
SUBJ [ PRED 'SANBAI' ]- - - - . PRED 'EAT <SUBJ OBJ>' : :
: :
OBJ POSS PRED PRO - - - ' : REFL - - - - - ' PRED 'VINEGAR' CL DE
118. San1bai2i chi1 ta1*i/j-de cu4. (idiomatic reading)
Sanbai is jealous of him.
-f
SUBJ [ PRED 'SANBAI' ]
PRED 'BE-JEALOUS <SUBJ OBL>' : :
OBL PRED PRO : REFL - - - - ' CL DE VFORM CHI
The non-reflexive pronoun ta1 in Chinese has the binding attribute [-ncl], meaning that it must not find an antecedent within the minimal nucleus. A nucleus is in turn defined as an f-structure containing a PRED attribute. Within 117-f of the regular possessor, the non-reflexive pronoun thus must not find an antecedent within OBJ, the minimal nucleus ta1 is in, and is therefore free to be bound with Sanbai the matrix subject or some other element outside of the sentence. Within the idiomatic f-structure 118-f, however, the minimal nucleus that contains the non-reflexive pronoun is the entire f-structure; thus, as predicted, ta1 cannot refer to Sanbai the subject. Huang's account thus seems adequate concerning these data. Unfortunately however, behavior of the non-reflexive pronoun is not so straightforward. Consider first the following regular verb phrase, which shares the POSS f-structure with 119-f above.
119. Li3si4i ying2de2 le ta1*i/j-de xin4ren4.
Lee win PERF s/he POSS trust Lee won his trust.
-f
SUBJ [ PRED 'LEE' ] PRED 'WIN <SUBJ OBJ>' : :
OBJ POSS PRED PRO : REFL - - - - - ' PRED 'TRUST'
Unlike 117-f, the non-reflexive pronoun here cannot be bound with the matrix subject, while Huang's account predicts that it would since SUBJ is outside the minimal nucleus containing the pronoun. The next sentence presents a problem that is quite the opposite.
120. Li3si4i hen3 hou4hui3 ta1i/*j-de xing2wei2.
Lee very regret s/he POSS behavior Lee regrets his behavior.
-f
SUBJ [ PRED 'LEE' ]- - - -. ADJ { [ PRED 'VERY' ] } : PRED 'REGRET <SUBJ OBJ>' : :
OBJ POSS PRED PRO - - - ' REFL -
PRED 'BEHAVIOR'
121. Li3si4i hen3 hou4hui3 zi4ji3i/*j-de xing2wei2.
Lee very regret self POSS behavior Lee regrets his own behavior.
Again, while f-structure 120-f here is identical to 117-f of regular POSS and to 119-f, it does not allow the non-reflexive pronoun to be free outside of the scope of the sentence. Instead, the pronoun must be bound with the matrix subject and nothing else, again contrary to the prediction of Huang's account. In other words, the verb hou4hui3 'regret' forces the non-reflexive pronoun (ta1 in 120) in its OBJ to behave like a reflexive pronoun (zi4ji3 in 121).
Since the general pattern of non-reflexive binding in Chinese is indeed as described in Huang's account and represents the unmarked case (C. Huang 1990a:286), the deviations, or the marked cases, in 119 and 120 would have to be lexically sanctioned and override the general binding principles. Within such an analysis, it is entirely unnecessary to pose a different f-structure with OBL for the possessive NP1 in idiom chunks.
In 119, the binding pattern of the possessive NP is sanctioned by the head noun xin4ren4 'to trust', a nominalized transitive verb, which specifies that its subjective [REFL -] POSS be free within the minimal nucleus containing the POSS and a SUBJective function; the head verb
hou4hui3 'to regret', on the other hand, requires that the subjective POSS in its OBJ be [+ncl +sb],
i.e., bound with the SUBJ of the minimal nucleus that contains the POSS and the SUBJ.
This analysis of lexically sanctioned binding relations (which can override the general binding principles) thus accounts for all the grammatical sentences in 122-123, and also provides a sound explanation for the ill-formed 124, where the joint binding requirements from the object
xin4ren4 'to trust' and the verb hou4hui3 'to regret' force ta1 to be bound simultaneously with both
the matrix SUBJ and some other element outside the sentence; consequently, the sentence has an impossible binding relation.
122a. Li3si4i zheng1qu3 ta1*i/j-de xin4ren4.
Lee strives for his trust.
b. Li3si4i xu1yao4 ta1*i/j-de xin4ren4.
Lee needs his trust.
c. Li3si4i zai4hu1 ta1*i/j-de xin4ren4.
Lee cares about his trust.
d. Li3si4i gu1fu4 le ta1*i/j-de xin4ren4.
Lee betrayed his trust.
123a. Li3si4i hen3 hou4hui3 ta1i/*j-de lu3mang3.
b. Li3si4i hen3 hou4hui3 ta1i/*j-de tan2hua4.
Lee regrets his talk.
c. Li3si4i hen3 hou4hui3 ta1i/*j-de zuo4fa3.
Lee regrets his method.
d. Li3si4i hen3 hou4hui3 ta1i/*j-de cu1xin1.
Lee regrets his carelessness.
124a.*Li3si4i hen3 hou4hui3 ta1ij-de xin4ren4.
In light of the independently needed lexically sanctioned binding relations, I thus contend that the possessive NP in VO idioms has exactly the same f-structure as the regular literal possessive NP and that the 'exceptional', marked, binding relation that the possessive NP1 exhibits in idioms is lexically sanctioned by the individual idioms. Another way to look at this is that the idiomatic reading is obtainable only if the [REFL -] POSS, if any, is not bound with the SUBJ in the minimal nucleus containing a SUBJ.
125. San1bai2i chi1 ta1i-de cu4. (literal only)
-f
SUBJ [ PRED 'SANBAI' ]- - - -. PRED 'CHI <SUBJ OBJ>' : :
OBJ POSS PRED PRO - - -' REFL - PRED 'CU4' CL DE
126. San1bai2i chi1 ta1j-de cu4. (literal and idiomatic)
-f
SUBJ [ PRED 'SANBAI' ] PRED 'CHI <SUBJ OBJ>' : :
OBJ POSS PRED PRO : REFL - - - - - ' PRED 'CU4' CL DE
This analysis is further confirmed by the fact that a non-reflexive possessive pronoun in some VO idioms actually behaves exactly opposite and requires a binding relation with the SUBJ of the minimal nucleus, again similarly to the reflexive pronoun zi4ji3.
127. Ta1i zuo3 ta1i/*j-de Yang2guan1 dao4.
s/he go s/he POSS Yangguan road He goes his own way.
128. Li3si4i ying4 zhe ta1i/*j-de tou2pi2 cheng2ren4 le.
Lee harden PROG s/he POSS scalp confess PTCL Lee forced himself and confessed.
129. The Presidenti again ate hisi/*j words.
130. The Presidenti is talking through hisi/*j hat again.
131. The Presidenti put hisi/*j foot in hisi/*j mouth again.
Huang's account of the so-called POBJ with an OBL function cannot cope with the data cited above, where the non-reflexive possessive pronoun, again behaving rather like a reflexive pronoun, must be bound with the SUBJ within the minimal nucleus that contains itself and a SUBJ. Nor can his account explain why a idiom may require a reflexive possessive pronoun and forbids a non-reflexive one.
132. Li3si4i da3 zi4ji3i-de zui3ba1. (=,#)
Lee slap self POSS mouth a. Lee slapped his own mouth. b. Lee contradicted himself. 133. Li3si4i da3 ta1i/j-de zui3ba1. (=)
a. Leei slapped hisj mouth.
b. Leei slapped his owni mouth.
The idiomatic reading is obtainable only when the required reflexive possessive pronoun is present (132). More interestingly, even when a non-reflexive possessive pronoun is bound with the matrix SUBJ, the idiomatic reading is not obtainable (133). All these examples clearly demonstrate that specific syntactic requirements must be fulfilled, including binding relations, for a idiomatic reading to obtain, and more importantly, that it is entirely unnecessary for the idiomatic reading to have a different f-structure from that of the literal reading.
To summarize, for each supporting argument that Huang has raised, the alternative that does not involve a distinct f-structure is found to be consistently more viable once further data is considered. Furthermore, Huang's f-structure account has two major drawbacks. First, the account entails that for each type of idiom with a different c-structure, two c-structure rules are needed, one for the regular phrase and the other for the idiom phrase. The two rules are identical in their c-structure components but different in their functional schemata (for the building of different f-structures). This would cause the proliferation, if not doubling, of c-structure rules; after all, idioms come in numerous different constituent structures and the potential is unlimited. Secondly, whether this account assumes the current lexical mapping theory or the pre-1986 'classical' model of lexical rules, it does not account for morpholexical processes, e.g., passive, dative, locative inversion, that involve an element in the idiom as a thematic argument or subcategorized function.
4. THE THEMATIC STRUCTURE SOLUTION
Given the fact that syntax proper, i.e., the c-structure and the f-structure, offers no adequate solutions to the ambiguity of idiom chunks, the next logical place to explore is the thematic structure. This brings us to the 'classical' LFG treatment of idiom chunks proposed in Bresnan (1982b).
Bresnan's solution assigns different thematic structures to the lexical heads of the idiomatic reading and the literal reading. I will discuss extractable idioms like keep tabs on first. The lexical
head keep is a three-place predicate, requiring thematic structure <ag th loc>, for its literal reading. In the idiomatic reading, however, it is a two-place predicate, requiring <ag th>, similar to verbs such as watch or investigate that require two arguments, and the assignment of grammatical functions is the following (Bresnan 1982b:46):
134.
Mary kept money on the counter. keep <ag th loc> 'thematic structure
<S O OBL> 'subcategorization 135.
The FBI kept tabs on John.
keep-tabs-on <ag th> 'thematic structure <S OBL> OBJ FORM TABS 'subcategorization
Note that this treatment is formulated within the so-called classic, i.e., pre-LMT, model of LFG, where function-changing rules such as passive and dative were stated in function terms. For example, the 'classical' passive rule converts OBJ to SUBJ, and suppresses the SUBJ or converts it to an OBL. (Rule repeated in 136.)
136. Passive (Bresnan 1982b): SUBJ /OBL OBJ SUBJ
Within this formulation, the passive counterpart of the idiom, shown in 137, is accounted for, where tabs, the OBJ, is now the passivized SUBJ. Moreover, the ungrammatical 138 is ruled out, for the theme role, John, as an OBL, cannot be passivized.
137. Tabs were kept on John (by the FBI). 138.*John was kept tabs on (by the FBI).
Nonetheless, as I have demonstrated in Chapter V and VI, the mapping from the thematic semantical structure to the so-called lexical form is no longer stipulated in the current LFG framework, rather it is principled and constrained by the various well-attested lexical mapping principles. As convincingly argued in Levin (1987) and Bresnan and Kanerva (1989), among others, the lexical mapping theory, one of the most important theoretical advances in LFG in recent years, constitutes a tremendous improvement over the arbitrary stipulations in the 'classical' mapping between thematic roles and grammatical functions. Within LMT and its formulation of passive, however, Bresnan's treatment of idiom chunks quickly runs into difficulty. As shown below, while LMT correctly predicts the lexical form of keep in its literal reading (139), it incorrectly links theme, i.e., John in 140, of the idiom keep tabs on to OBJ, rather than the on-marked OBL required. To be fair, I will assume the version of LMT in Bresnan (1989) and Bresnan and Kanerva (1989) (see 3.1.1, Chapter V for details).
139. Mary money on the counter keep <ag th loc>
IC -o -r -o DC -r +r --- WF <S O OBL> 140. FBI John tabs keep-tabs-on <ag th>
IC -o -r DC -r --- S S/O WF S *O
This account predicts that <ag th> maps to the incorrect <SUBJ OBJ>, rather than the well-formed <SUBJ OBL>. The outcome is no more encouraging when we take into consideration the non-thematic OBJ stipulated in the lexical form of keep-tabs-on, whose FORM is designated to be TABS. Given the Function-Argument Biuniqueness Condition, the theme role cannot map to any grammatical function, as shown below, since the OBJ function is no longer available. There is simply no way to derive the required lexical form, <SUBJ OBL> OBJ FORM TABS, without resorting to the earlier arbitrary stipulations of grammatical functions.
141. FBI John
keep-tabs-on <ag th> OBJ FORM TABS IC -o -r
DC -r
--- S S/O OBJ FORM TABS WF <S ?> OBJ FORM TABS
Worse still, given the passive rule within LMT (repeated in 142), the well-formed passivized counterpart of the idiom (137) cannot be accounted for, while the ill-formed passivized theme (138) is incorrectly predicted to be grammatical.
142. Passive (BK 1989): < ..>
143. FBI John
keep <ag th> OBJ FORM TABS IC -o -r
PASSIVE DC
--- WF <S> OBJ FORM TABS
144.*John was kept tabs on (by the FBI).
One might suggest that keep-tabs-on be treated as a lexical unit to get around the above-mentioned problems. Modification and extraction, among others, provided two kinds of evidence against this proposal, see 145-146.
145. The FBI is keeping close tabs on John. 146. Close tabs are being kept on John.
In 145, tabs can be modified by an adjective and it can be passivized as well (146). Both
sentences demonstrate that keep tabs on violates lexical integrity and is thus a phrase. It is quite clear then, that the lexical mapping theory dictates that an element accessible to any morpholexical rule, such as passive or dative, must be thematic. In this idiom, tabs can be modified as well as passivized; thus, it must bear a thematic role. Rendering a thematic element non-thematic, Bresnan's (1982b) classical treatment of idioms like keep tabs on is bound to fail within LMT.
Therefore, for extractable idioms, idiomatic and literal readings must share the same thematic structure.
Bresnan's account fares much better for non-extractable idioms, such as kick the bucket. While kick has thematic structure <ag th> for its literal reading, the idiom has <th> for its meaning of 'to die'. Lexical mapping also predicts the two lexical forms correctly.
147. John kicked the barrel.
kick <ag th> IC -o -r DC -r --- S S/O WF <S O>
148. John kicked the bucket. (meaning 'John died')
kick-the-bucket <th> OBJ FORM BUCKET IC -r
DC
--- S/O
WF <S> OBJ FORM BUCKET
The primary reason for the thematic structure <th> is of course because kick the bucket has the idiomatic reading of 'to die', a one-place predicate. Therefore, the bucket, or more precisely the grammatical function OBJ subcategorized for by kick, must be stipulated to be non-thematic. However, consider the idiom rock the boat, which, though non-extractable just like kick the bucket, cannot be paraphrased in a one-place predicate; rather, it has to be understood as something like 'to disturb the situation', a paraphrase syntactically parallel to the idiom itself. Therefore, rock can have a single thematic structure <ag th> for its literal reading as well as the idiomatic reading, as long as OBJ FORM ROCK is specified in the lexical form. Likewise, the idiom kick the bucket can just be easily understood as 'to lose (one's) life', which, compared with 'to die', is syntactically much more comparable to kick the bucket. If so understood, kick would be the same two-place predicate in both the idiomatic reading and the literal reading.
For verbal idioms with a subcategorized PP, e.g., fall on deaf ears, disappear into thin air, and throw one's hat in the air, Bresnan's account would again have problems. It would not have a thematic role for idiomatic thin air for example, as shown in 149 below, which must be stipulated in the lexical form, much the same way [OBJ FORM BUCKET] is stipulated for kick the bucket.
149. The public support has disappeared into thin air. (#)
disappear-into-thin-air <th> OBLloc FORM AIR
IC -r DC
--- S/O
WF <S> OBLloc FORM AIR
150. The UFO has disappeared into the dark sky. (=)
kick <th loc> IC -o -o DC -r +r --- S OBLloc WF <S OBLloc>
In its non-idiomatic use, as in 150, disappear has the thematic structure <th loc>, which, via lexical mapping, links to lexical form <SUBJ OBLloc>. However, not allowing the idiomatic into
thin air to be linked to a thematic role, this account violates a universal condition on grammatical
functions as specified in the lexical mapping theory (e.g., Bresnan and Kanerva 1989, or see Chapter V for details). Recall that SUBJ and OBJ are the only two functions that are classified as [-r], i.e., non-restricted to any particular thematic role; thus, only SUBJ and OBJ are allowed to be non-thematic, i.e., bearing no thematic role. An OBL function, however, is classified [+r], or thematically restricted, and must be linked to a particular thematic role. Therefore into thin air as OBLloc must be linked to a locative role in the a-structure. The analysis in 149 is therefore
impossible given the current theory of lexical mapping. Furthermore, this analysis also fails to account for the possible locative inversion constructions, as in 151-152. In general this account cannot handle idioms that allow function-changing constructions that in LMT terms must involve thematic roles, such as passive, locative inversion, and dative shift.
151. However, soon after the election the scandal exploded and into thin air disappeared the public support. 152. In spite of all the signs of Iraqi aggression, on Washington's deaf ears fell his repeated warnings.
In short, Bresnan's account, though an alternative for non-extractable idioms, is unworkable for idioms that undergo relation-changing morpholexical operations or subcategorize for an oblique function, which, by definition, is thematically restricted and thus must be linked to a specific thematic role. This account therefore does not provide a consistent explanation to all types of idioms.
5. THE LEXICALIZED METAPHOR SOLUTION
The conclusion to be derived from the discussions above is this: an idiom is either fixed or syntactically analyzable, and in the former case, it should be considered a single lexical item with lexical integrity, while in the latter case, the thematic structure of the head predicator and the f- and c-structures of the idiom are no different from those of the regular, literal expression. To account for the idiomatic readings, I propose a solution based on Her, Higginbotham, and Pentheroudakis (1994), Wasow et al (1983), and Lakoff (1987); a solution that considers the subparts of an idiom as analyzable and compositional elements with metaphorical references.
Her et al (1994) present a treatment of idioms within an LFG-based machine translation system and recognize two types of idioms: 1) lexical idioms, continuous phrases stored as lexical units in the lexicon; 2) transfer idioms, idiom phrases that parse compositionally to produce a non-ambiguous f-structure but translate literally and idiomatically to two distinct phrases in the target language. In this approach, a lexical idiom is a fixed expression, or more precisely a lexical item, for example you-know-who, who-done-it, tongue-in-cheek, or the many VO compounds that we have discussed in Chapter III and IV. A genuine idiom phrase must be syntactically analyzable and its syntactic structure is never ambiguous. Since I have demonstrated in earlier sections that previous accounts that pose distinct thematic structures, f-structures, or c-structures for idiom phrases are all unsatisfactory or unworkable, the solution I propose will adopt Her et al's position that idioms do not have different syntactic or thematic structures.
Wasow et al (1983) examine several different types of idioms in English, most of which are of the VO construction and their conclusion can be seen as three closely related but logically separate claims. First, they argue that syntactically phrasal idioms have normal structures. This is
the same position upheld in Her et al (1994), Stock (1987), Wasow et al (1983), Gazdar et al (1985), among others, and it is also the position I will take here. Their second claim is that, semantically, idioms differ in terms of whether the assignment of the idiomatic meaning is to the idiom chunk as a whole or to its parts. In other words, some idioms are semantically analyzable and compositional, while others are not. They distinguish three classes: 1) noncompositional idioms (kick the bucket,
saw log); 2) conventionalized metaphors (take advantage of, spill the beans); and 3) compositional
idioms (pull strings). This position is also rather reasonable as that the associations between forms and meanings in idioms are not equally motivated or arbitrary. However, I do not agree with their conclusion that a uniform analysis of idioms is therefore not available.
The third claim that Wasow et al make is that the syntactic constraints of idioms are to a large extent predictable by the semantic relationships among their parts. While Wasow et al have made a strong case for their first two claims with ample examples and discussions, this last claim is much less substantiated and in fact may be logically circular. Take kick the bucket for example. They first claim that the idiomatic meaning is assigned to the whole phrase and not composed of idiomatic interpretations of the parts, and then claim that the fact that internal modifiers on the
bucket are ruled out is thus predictable because the bucket has no idiomatic meaning of its own.
However, I contend that the reversal of this argument is just sensible: that the bucket in the idiom does not allow modification or extraction indicates that it does not have an independent meaning. One can also argue that the idiomatic meaning of kick the bucket is composed of idiomatic interpretations of its parts and that this idiom imposes a set of syntactic constraints, including one that bars internal modifiers. After all, as shown in section 1 with the two verbs eat and devour and the several pairs of idioms kick someone's ass and kiss someone's ass, hold your horses and hold
your breath, and Chinese chi1 dou4fu3 (eat tofu) 'to flirt' and peng4 ding1zi (knock-against nail) 'to
be rejected', idioms with very similar semantic structures may behave in different ways syntactically. This line of argument also fares much better with the creative uses such as kick the
political/financial bucket; indeed, nothing in principle would rule out the possibility that these
creative uses become part of the conventionalized idiom.
Lakoff (1987) provides an account of idioms within the overall scheme of metaphor, metonymy, and conventional image, which is much more general. Here is a quote from him regarding the analysis on Japanese classifiers, which, I think, applies quite well to his analysis of idioms as well.
Ideally, each instance of a classifier outside the central sense should have a motivation. The motivation cannot be ad hoc--one cannot just make up a metony-my or image schema just to handle that case. This imposes a criteria of adequacy on the analysis of classifier languages. (Lakoff (1987:107)
Thus, each idiom should ideally have a motivation, be it metaphorical like blow one's top or
ji1dan4 li3 tiao1 gu3tou2 (egg inside pick bone) 'to be unreasonably critical', metonymical like put in a good word for someone or bi4 yan3 (close eyes) 'to die', mental imagery like saw logs or qiao4 bian4zi (stick-up braid) 'to die', or some types of combination of two or more motivations, for
example zheng1 yi1 zhi1 yan3, bi4 yi1 zhi1 yan3 (open one eye, close one eye) 'to turn a blind eye' can be seen as both metaphorical and metonymical and hell freezes over can be both metaphorical and imageable. I also agree that the motivation of any idiom should not be ad hoc; however, that does not mean motivations of idioms are all equally transparent. To use an example from Wasow et al (1983:111): saw logs and kick the bucket.
What is different about these two idioms, however, is that the relationship between the literal and idiomatic interpretations in saw logs is relatively transparent, viz., the sound of sawing logs is similar to that of snoring. Hence this
idiom is probably interpretable to those unfamiliar with it, by means of the normal mechanisms for interpreting metaphors (whatever they might be); in contrast, we presume that kick the bucket would be uninterpretable (on its idiomatic sense) to a first-time hearer.
Recognizing Lakoff's position and Wasow et al's observation, the point I want to make is this: the degree of motivation and the degree of transparency of the motivation may vary from idiom to idiom, and more importantly, from time to time and from speaker to speaker. For example, the seemingly unmotivated and opaque idiom kick the bucket, assuming its motivation was from the mental image that someone completes the act of hanging oneself by kicking the bucket one stands on, could be quite transparent at one time to speakers in certain regions and maybe even today. On the other hand, the idiom saw logs may still be part of a person's 'vocabulary' even if that person has never actually heard the sound of sawing logs. Much like the difference between etymology and folk etymologies (such as instances of back formation), it is a linguist's job to find out the overall network of systematic metaphors in a language (and perhaps culture) that idioms fit in and the history of each idiom. However, for the naive native speaker, many of the idioms may simply be conventionalized expressions and/or have motivations different from the genuine original motivations. Between opaqueness and transparency is thus a continuum.
As Lakoff has recognized (1987:451), idioms that have ambiguous, or even conflicting, idiomatic interpretations serve as the most illuminating examples for the varying degree of transparency in idioms. Consider the idiom a rolling stone gathers no moss. Although the motivation is clearly metaphorical, there are two primary interpretations of this metaphor that are nearly opposite: 1) one cannot accumulate good things, e.g., wealth and status, if one does not stay in one place, and 2) one loses his freedom or vitality if one stays at a place too long. In the first reading, the moss is viewed as a good thing, while seen as something negative in the second reading. An example from Chinese is bi4 yan3 (close eyes) 'to die' and bi4bi yan3jing1 (close eyes) 'to take a nap'. With the delimitive aspect of the verb bi4 'to close' by way of reduplication, the second idiom does not follow the first idiom to mean 'to die a little'. With a similar motivation based on an image and perhaps metonymy, the two idioms however do not share a similar semantic content.
Perhaps more drastically than regular lexical items, idioms may also widen or narrows its semantic range or become obsolete as time goes. Take idiom chi1 dou4fu3 (eat tofu) 'to flirt' for example. The image of tofu takes after the fair supple flesh of a woman and thus the idiom was first used to refer to a man's taking advantage of a woman in a physical manner with sexual implications. However, the idiomatic interpretation is now much wider: both sexes may be at the giving end or the receiving end and the action may be physical or verbal. For some speakers, this idiom has been even further bleached of ths sexual connotation and simply means to tease someone.
To be brief, in the solution I propose, phrasal idioms have regular syntactic structures, or a-structure, f-a-structure, and c-structure in LFG terms, and have motivations based on metaphors, metonymies, or mental images, with a varying degree of transparency between the literal reading and the idiomatic interpretation. Therefore, the solution consists of two essential parts: (1) syntactic constraints on the idiom interpretation of an idiom, and (2) motivation of the idiom interpretation. I will propose a formulation of (1) within LFG and adopt Lakoff's treatment of (2). I will explain (2) first. Lakoff (1987:448) clearly defines the concept of motivation as follows:
The relationship between A and B is motivated just in case there is an independently existing link, L, such that A-L-B "fit together." L makes sense of the relationship between A and B.
