The Determinacy of Lexical Selectional Restriction and Its Basis for Existence

Dianzi University, China) Abstract: In this paper, the determinacy of regular lexical selectional restrictions is examined from both the internal structures of the single selectional restrictions (i.e. semantic selectional restrictions) and relationship between the structures of several selectional restrictions. Hence, our analysis and description are shifted from casual and indeterminate strings and markers on case basis to determined rules and circumspect dynamic systems, from lexical precepts to knowledge precepts, from the state of memory to rational deduction and rhetorical devices. Further studies indicate that selectional restrictions are intertwined structures, a feature that makes it possible to be one of the bases for lexical selectional restrictions to come into existence. Its related theories are the grarantee for scientific observations of


Introduction
Both syntax and lexicology study lexical relationship from different perspectives. Syntax focuses on the abstract syntactic relations between lexicons in syntagmatic structures, while lexicology on precepts of interlexical relations. Starting from lexical restriction relationship, a hot topic in both syntax and lexicology, this paper is to demonstrate the determinacy and ontological basis for regular lexical selectional restrictions (if we put aside pure paradigmatic relation and super selectional restriction relation) by the analytical studies on the relations between internal determinacy and external determinacy of semantic  Project: National Social Sciences Founding Project (China), "Study of Sentence Complexity in Modern Chinese" (No. 11BYY084).  Dr. Ma Qinghua: Professor, School of Liberal Arts, Nanjing University, Nanjing, 210046, China. E-mail: maqh86@ nju.edu.cn.  Huang Yanqun, Hangzhou Dianzi University, Hangzhou, 310018, China. E-mail: teresahyq@163.com. selectional restriction structures, and to shift our analysis and description from casual and indeterminate strings and markers on case basis to determined rules and circumspect dynamic systems, from lexical precepts to knowledge precepts, from the state of memory to rational deduction and rhetorical devices. Furthermore, links for selectional restriction are intertwined structures, which is one of the bases for linking lexical selectional restrictions. Its related theories are the guarantee for scientific examination of selectional restriction structures in the Chinese corpus.

The internal determinacy of selectional restriction structures
The premise for scientific research into lexical relationship is to collect all data for studying lexical association in an objective, confirmative and easy way. In previous studies, whether in introspection or in survey, one word was chosen as the stimulus to retrieve its response words. It was not until their universality of psychological relations was estimated that the associative relations between them can be claimed. While the basic selectional restriction structure consists of a given word, a selectional restriction word and their relationship, the non-restrictive stimulus-response model that is conventionally adopted in the academic circle only takes the units into consideration, ignoring the relationship between units, as a result of which the data are considerably random and fall short of scientific description. Without semantic restriction, the number of response words is beyond measure which can be better shown in the stimulus-response model of noun, for example, zhēn(针, "needle") as the trigger. To the contrary, the addition of the semantic relationship between the trigger and the unknown response word is conducive to the conclusion that response words can be the relatively target words. Compared with the non-restrictive stimulus-response model, this kind of restrictive stimulus model makes for objectivity and science with its substantially less random data.
1.1 The concept of determinate selectional restriction structure Surely, the restrictive stimulus model is relatively limited in promoting the determinacy degree of word associations. Even if this model is adopted, the determinacy degree of word associations still varies from word to word, and from context to context. The concept of semantic selectional restrictions can be classified into the following types based on its scale: a. a determinate concept as a word or a group of synonyms (Ma Qinghua, 2000:70) such as zuǐ/kǒu(嘴/口, "mouth")in example 1; b. the multiple closed or quasi-closed one, see example 2; c. semi-closed one, see example 3; d. the open, see example 4. The determinacy of selectional restriction structures depends on individual component in the selectional restriction set. The more potential selectional restriction concepts in the set, the more impossible it is to predict individual component. It is obvious that a-type selectional restriction concept is of highest determinacy while others of lower determinacy and higher indeterminacy. Members in b-type selectional restriction set cannot be activated as determinate as a-type but every concept is of high approachability and predictability. From type b to type c, and to type d, the indeterminacy in selectional restriction concepts is characterized by their ascending order.
This concept seems to be tailored for the constraint of a given word, for constraint being the heterogeneous relationship or conceptual overlapping between lexical meanings (Ma Qinghua, 2005:2), the given word gives specific requirements for the characteristics of selectional restriction concept as if making a slot or blank and the sememe reflecting the requirement is represented by feminine sememe (marked by [♀/]), for example, the feminine sememe for chī (吃, "eat") being [♀/+mouth]); the selectional restriction concept fills the slot or blank in compliance with the requirement and the sememe reflecting the filling requirement is represented by masculine sememe (marked by [♂/]) , for example, the masculine sememe for zuǐ (嘴, "mouth") being [♂/+mouth]. The feminine sememe of a given word and the masculine sememe of saturated concept are identical with each other in content, just like replicas. ① Three levels of saturated selectional restriction concept are specific saturated selectional restriction concept without the superordinate (e.g. 6a), general saturated selectional restriction concept about high-level abstract concept and the intermediate saturated selectional restriction concept, like 7a. Naturally, the formation of saturated selectional restriction concept is caused by semantics and structure. Semantically speaking, the specific saturated selectional restriction concept can describe things in a specific and expressive way (e.g. 6b); the intermediate one can intensify or disprove abnormal cases ① Empty sign(空符号) and full sign(满符号) are borrowed from Gordon Pask(戈登· 帕斯克) and Susan Karan(苏珊· 卡冉) (1987). Xing Gongwan(1983:19)considers lexical collocation as complementary, but does not differentiate the correlated sememes, limiting the further study of the problem.
(e.g. 7a); the general one can describe things in a generalized way.

The unidirectional dependence of the saturated structure
In the restrictive stimulus-response model, if a saturated selectional restriction concept B can be obtained from the given word A whereas a given word B cannot get a saturated selectional restriction concept A, then the unbalanced relationship between A and B is called unidirectional dependence (Ma Qinghua, 2000:219). Comparing 10a and 10b: (10) a.[duǒ(朵, "a measure word for flowers")] object measured →huā(花, "flower"); b.[huā(花, "flower")] unit of measurement →{ duǒ,shù,pén(朵, 束, 盆, "buddle, bunch, pot, etc.")} Of the saturated structure of unidirectional dependence, some saturated selectional restriction concept functions as the core of the syntactic meaning while others not. Those as the core, like the saturated selectional restriction verb in VRS (e.g. ex.Ⅰ below) and the noun of saturated selectional restrictions in the noun-quantifier collocation (e.g. ex.Ⅱ below), are quite small-scaled compared with the same type of non-saturated selectional restriction structure. Wang Yannong & Jiao Qun & Pang Yong (1987) made a research into the collocation of verbs with 322 adjectives and verbs acting as resultative complements and with verbs. On the basis of the result, the conclusion after checking and calculating is drawn that there are only 2 cases of saturated structure (e.g. ex.Ⅰbelow) and that saturated structure with verb as the core of the saturated structure of selectional restriction concepts is 1/161 times the number of the non-saturated structure of its kind. Liu Xuemin & Deng Chongmo (1989) examined over 8,000 nouns and 388 collocative quantifiers, of which we obtain 59 nouns that can be saturated selectional restriction concept and the other 329 nouns can only be non-saturated restriction concepts and the proportion between the saturated noun and the non-saturated is 1:5.6 through checking and calculating. Of the two structures, there"s no chance for complementary predicate and quantifier to be selectional restriction concept.

The bidirectional dependent saturated structure
In the restrictive stimulus-response model, if saturated selectional restriction concept B can be obtained from the given A and saturated selectional restriction concept A can also be obtained from the given B, then the interdependence between A and B is called bidirectional dependence, equal to a circuit composed of two unidirectional dependence cases of different direction and type. Such can be shown in the saturated structure of onomatopoeia-noun and a few predicate-case relations, e.g. a and b in 12 and 13. Its units are saturated selectional restriction concepts to each other but the semantic types of constraint are different. B in 13 is at most quasi-saturated structure because instrument can act as something with fixed purpose or provisional function; for example, "the instrument" usually is used to "play" and to "beat or smack" as an attack provisionally. The bidirectional construct is much fewer than the unidirectional from the viewpoint of the total sum of numbers and internal scales and saturated structure fewer than the non-saturated in language. It is determined by the linguistic information theory principle that there is more chance for two random words to belong to a non-saturated structure than a saturated structure, because a saturated structure is a redundant and kind of "poorly used" structure except for emphasis. It is the fact that the saturated selectional restriction concept is the induction of the much larger amount of non-saturated selectional restriction concepts, e.g. the non-saturated structure Xiǎomí ng chī(小明吃, "Xiaoming eats") | māma chī(妈 妈吃, "mum eats") | gǒugou chī(狗狗吃, "the dog eats")| xiǎojī chī(小鸡吃, "the chick eats") etc. and the saturated structure dòngwù chī(动物吃, "animals eat").

Saturated selectional restriction meaning cluster
When non-saturated selectional restriction concept is of such high indeterminacy that the set it is in becomes open, all concepts in the set have to be governed by a meaning cluster (i.e. the sememes formed by conjunction). Otherwise, the set is beyond control. Therefore, the psychologically indeterminate selectional restriction association becomes descriptively determinate but what is determinate concerns different object and way. The meaning cluster can be called "saturated selectional restriction meaning cluster", e.g. the open selectional restriction concept set in 14a and selectional restriction meaning cluster in 14b.
[qiāo(敲, "knock")] object acted upon →[+object+hard shell] Words referred to by the saturated selectional restriction meaning cluster form a lexical gap in the vocabulary system. The meaning cluster of lexical vacancy stems from induction from a certain number of non-saturated selectional restriction concepts or words (e.g. 14a) and its meaning is the same as that of a phrase (e.g. the meaning cluster in 14b is the same as phrase "object with hard shell".). It produces from the subconscious level, for example, a normal native speaker can generalize the mode of "qiāo(敲, "knock")+[X]object" with qiāo(敲, "knock") as the invariant from qiāo zhuōzi (敲桌子, "knock the desk")|qiāo hēibǎn (敲黑板, "knoch the blackboard")|qiāo tāde tóu (敲他的头, "knock his head")|qiāoqiāo qiá ng(敲敲墙, "knock the wall") etc. in different contexts based on the commonality seeking operation and subconsciously postulates that the conceptual value for X is "object with hard shell". However, there is no lexical item for X, so only the masculine meaning cluster [♂/+ object with hard shell] can directly govern all potential variants of X, like non-saturated selectional restriction objects zhuōzi( 桌子, "desk") | hēibǎn( 黑板, "blackboard") | tāde tóu(他的头, "his head") | qiáng(墙, "wall"), etc. and all potential variants can only be captured by the object of the masculine meaning cluster [♀/+object+hard shell]of chī(吃, "eat") and the feminine meaning cluster is an ideology as a complex concept to reflect the recipient requirement and wait to be tenonned.
Meaning cluster is no fixed concept formed by words, inappropriate to act as the starting point of stimulus units or structural description in the restrictive stimulus-model. Thus structures including saturated selectional restriction meaning cluster can only be considered as unidirectional dependence, supporting the argument that unidirectional dependence has an advantage over bidirectional dependence.
1.4 Candidate concept in multiple closed or semi-closed type If a given word is faced with a selectional restriction object of a multiple closed or semi-closed set in restrictive stimulus-response model, to choose and utilize every concept is indeterminate to some degree but it is not absolutely beyond prediction due to the closed-nature of the set. In the context selectional restriction association can be inferred from a specific word (e.g. 2 and 3) by the selectional restriction structure (including language inference and common sense, etc.) while it is different for an abstract word. The more abstract the given word, the more relevant synonyms (similar but different in constraint range), the more fuzzy the boundary of its connotative meaning, and as a result, the more difficult it is to carry out its deduction. Therefore, foreign students often make errors. Statistics about the prone-mistakes committed by both foreign and Chinese students in The Contemporary Chinese Mistakes and Correction Dictionary compiled by Yang Qinghui show that synonymy mistake account for 47% of the double sound word mistakes (Zhu Zhiping, 2004), most of which are abstract synonymies (e.g. 15 and 16) without regular connotation and boundary. Thus it is better to list the candidate concepts in selectional restriction set for plain explanation. For example, in Zhang Zhiyi(1981:65, 150) and Dictionary Compilation Bureau of the Language Research Institute of Chinese Academy of Social Sciences (2005), yōuliáng(优良,"good") is used to modify personality, working style, skill, species, quality, tradition, condition and environment, while yōuxiù(优 秀,"excellent") to modify people with high education and good quality, and man-made products, especially spiritual achievement, like staff, role or his knowledge, works, accomplishment and heritage; xuānbù(宣布,"announce") can take nouns like fǎlìng(法 令,"law"), jìlǜ(纪律,"discipline"), zuì xí ng(罪行,"crime"), míngdān(名单,"list"), jiéguǒ(结 果,"result") as its objects, and some verbs and clauses like qì quá n(弃权, "abstain"), tuìchǎng(退场, "exit"), ché nglì (成立, "found"), jiēshòu yìjiàn(接受意见, "follow one"s advice"), guójiā dànshēng(国家诞生, "a nationis founded"), mì nglì ng(命令, "command"), xīnwén(新闻, "news") etc. can collocate with it, too.

The determinate association between selectional restriction structures
Several selectional restriction structures are linked up through such methods as language, common sense and rhetoric in an organic, even determinate way instead of in a disordered way. Therefore, based on his semantic experience, one can consciously expand his application by following a certain way and the determinacy doubtlessly is the cornerstone of this expansion. Xing Gongwan (1983:21) says, "It is difficult to understand the coexistence and collocation of each content word together with other content words in the same segment." However, another argument, "Lexical constraint can only be acquired by training in word and phrase collocation instead of inferring on a large scale like the morphological change" (Li Rulong and Yang Jichun, 2004) is questionable. If saturated structure and common sense are a large sum deposited in a bank, the potential entailment choice is like free withdrawal from its part deposit. Below are the three means of constituting deterministic links.
2.1 Linguistic means Linguistic inference can be conducted by bridging the precepts of lexical paradigmatic association (mainly logic relation, like hyponymy) between the selectional restriction structures. The linguistic reference of selectional restriction structure means that selectional restriction structure B can be proved by inferring from the rightness of structure A on the basis of linguistic association. The logic relations between lexical paradigmatic associations can be classified as hyponymy "niǎo(鸟,"bird") -má què (麻雀,"sparrow")", polar opposites "zà nché ng(赞成,"agree") -fǎnduì(反对,"disagree")", relational opposites "jiāo( 教 ,"teach") -xué ( 学 ,"learn")", synonymy "xìxiǎo( 细小,"tiny") -wēixiǎo( 微 小,"little")", part -whole "niǎo(鸟,"bird") -chìbǎng(翅膀,"wing")" and synergy "lí zi(梨 子,"pear") -píngguǒ(苹果,"apple")", etc ( In the course of inference, the selectional restriction concept is replaced by its hyponyms, most of which are nouns. There are three reasons for this: firstly, the number of nouns is far greater than that of predicates and other closed lexical items, which is universal in all languages; secondly, the sum of any noun"s selectional restriction scale by collocation is enormously huge; lastly, the noun"s application of common sense is more direct and large-scaled and the common sense could be directly drawn from abstraction and description of real objects instead of induction of precepts for lexical association in language (e.g. below). Therefore, nouns abound with hyponymy words which are available in collocation and replacement, like noun is replaced in 17 as a successful inference and verb is replaced in 18 as a failing inference.

Application of common sense
The analysis of lexical collocation involves common sense; illustrating meaning and syntax also shows its systematic insufficiency when it comes to be involved with common sense. (Ma Qinghua, 2005:414-416) Common sense refers to the routine semantic knowledge in the knowledge system and there exists unidirectional dependence on common sense of selectional restriction structure, at least concerning two relational models.

Natural linkage selectional restriction by procedural knowledge
The daily procedural knowledge is a kind of common sense for linage of several declarative relationships (Ma Qinghua,2000:250). The saturated structures or non-saturated structures involved with predicate can be naturally lined up on the basis of procedural knowledge and all these selectional restriction structures correspond to an event. (e.g. 9b and 4-5)

Inference with common sense as the insignificant premise
Sometimes a new selectional restriction structure can be inferred by means of certain common sense, this kind of inference between the extralinguistic world and linguistic system is called "common sense and language inference".
Meaning cluster can expand the scale for the determinate association between selectional restriction structures. Meaning cluster is no fixed concept expressed by words, and in a restrictive stimulus-response model, it is unsuitable to be a stimulus unit or the starting point of structural description. However, it can act as transition in or insignificant premise for the inference between selectional restriction constructs by using common sense concerning relevant nouns instead of linguistic knowledge. Of all parts of speech, the application of common sense by nouns is the most direct and abundant and the common sense is about the abstraction and description of reality characteristics instead of induction of precepts of lexical association in language. The selectional restriction concepts of nouns are of various kinds. The definition of a noun includes some common sense component, which corresponds to a statement with regard to meaning(Ma Qinghua, 2000:255), and it contains all basic original information of lexical collocation and the original selectional restriction structures, available as the starting point of semantic inference. The definition of the noun zhēn(针, "neddle") in Contemporary Chinese Dictionary is like 25a. We can capture its concept directly from reality without relying on the linguistic induction. The aforementioned common sense concerning zhēn(针, "neddle") can be decomposed into common sense components, as described in 25b, in which some original selectional restriction structures are connotated (e.g. 25c).

Rhetorical devices
The rhetorical devices used in conventional selectional restriction collocation are conventionally limited and cannot be deviant, that is, excluding deviant collocation. Synecdoche and metonymy are conventional rhetorical devices in lexical usage, expanding the range of lexical selectional restriction linkage and building a bridge of meaning equivalence between the two selectional restriction structures before or after the rhetorical devices used.
Synecdoche means that part stands for the whole, the whole for part, the genera for the species or the species for the genera. The semantic inference through synecdoche between selectional restriction structures is conducted only in two ways: 1. part stands for the whole ( e.g. 28 and 29); 2. advance towards the superordinate concept of the original concept, that is, the category for the specific type ( e.g. 30 and 31). Thus, if the original concept is saturated selectional restriction concept, synecdoche makes the new selectional restriction structure ambiguous oversaturated structure. And the semantic determinate association may be lost by inverting the starting point, e.g. 31a cannot be inferred by 31b in 31. Metonymy is formed based on association, for example, contiguity, coexistence and cause-effect relation with such common types as the method standing for the subject, cause for the effect, material for the product and process for the product. Take the selectional restriction structure of metonymy as example: 32b is realized by adopting the metonymy of the product for the process. At best the general rules of formation for the links of selectional restricted structures formed by rhetorical devices can be generalized but in nature the structure belongs to the linguistic agreement category with little deduction, for example, in 31b the result chī(吃, "eat") recipient →wùzhì(物质, "matter") inferred by synecdoche is usually unacceptable and the latter not conventional selectional restriction association except for fun, speaking in a humorous way. 33b cannot be inferred from 33a through synecdoche.

The intertwined selectional restriction structure
A little more thought about the aforementioned analysis enables us to come to the conclusion that selectional restriction structure and theories about it can only be intertwine hierarchical structures, as the foundation for the existence of selectional restriction structure and its theoretical guarantee for scientific examination of selectional restriction structure. Strange loops emerge in the system of intertwine hierarchical structure as we go through certain level upward or downward the hierarchical system and find ourselves in the same place where we start our journey. All levels intertwine with and permeate, which imply infinite concepts (D. R. Hofstadter, 2001:15, 20). Zhou Guoguang (2002) and Qian Guanlian (2002) introduced Kurt Godel"s multilayer intertwine or strange loops theory into his examination of lexical problems directly or indirectly, but both are restricted to the explanations of the following linguistic common sense theories: the meaning of any word cannot explain itself and can only be explained by other words; other words cannot explain themselves either and can only be explained by other different words and the rest could be deduced from this. Actually, many intertwine structures hidden in the lexical selectional restriction association and its theories justify its existence and normal operation.

Intertwinement between the internal elements of selectional restriction structures
The internal elements of selectional restriction structures include the given word, selectional restriction elements and selectional restriction relation. The meaning intertwine and permeate between the given word and selectional restriction elements shows itself in the intermeshing between the feminine sememe of a given word and masculine sememe of the selectional restriction concept and between the feminine meaning cluster of a given word and masculine meaning cluster in the open selectional restriction set (mentioned before). The meaning intertwine between selectional restriction relation and selectional restriction units shows itself between the given word and the selectional restriction concept. It climbs upward the abstract steps and finally becomes a duplicate of the selectional restriction relation. (e.g. compare a and b in 34 and 35 below.)

Syntagmatic and paradigmatic intertwinement
The selectional restriction linkage at syntagmatic level cannot be thoroughly described without considering the lexical relation at the paradigmatic level. There are two reasons for this: firstly, the determinate relation between selectional restriction structures can only be established by means of certain relations on paradigmatic level. Secondly, the hierarchy of the conceptual classification at paradigmatic level is different from the abstract level required by selectional restrictions at syntagmatic level, but the two are closely related. The concept classification system on paradigmatic level usually has 5 or 6 levels, which are named according to scientific nomenclature from the abstract to the specific: most abstract name like zì rá n (自然, "nature"), abstract name like dìmào (地貌, "landscape"), less abstract name like shān(山, "mountain"), specific name like huǒshān(火山, "volcano"), more specific name like huó huǒshān(活火山, "active volcano"), and most specific name like āsūshān(阿苏山, "Asosan")(Ma Qinghua, 2000:26). Selectional restriction requirements on syntagmatic level do not correspond accurately to the hierarchical system but can be classified into different levels of abstraction with the saturated selectional restriction concept into specific, intermediate and general selectional restriction concept corresponding to the low, intermediate and high level with regard to abstraction. Li Rulong & Yang Jichun (2004) divided lexical selectional restriction linkage into semantic constraint and semantic collocation and they pointed out that: "Lexical constraint stems from semantic constraint and the restrictions of lexical collocation which seems to go beyond motivation, but to be governed by linguistic conventions." Indeed they classified lexical selectional restriction linkage into two levels, the semantic collocation relation with the selectional restriction linkage of specific word, and the semantic restrictive relation with the selectional restriction linkage of generic word or mid-lever word (a word that is somewhere between the specific word and generic word). Obviously, there is no point decomposing the selectional restriction requirements on syntagmatic levels into abstract levels if there is no conceptual classification hierarchy on paradigmatic level.

The multi-intertwinement of selectional restriction hierarchies among three levels of psychology, expression and description
Selectional restriction structure can be divided into psychological, expressive and descriptive levels: the restrictive stimulus-response model belongs to the psychological level (e.g. 34a), collocation in natural language flow the expressive level (e.g. 35a), and semantic component analysis the descriptive level or metalanguage level (e.g. 36a). The distinction among the three levels gives us a clear picture of the multi-tangled hierarchies of selectional restriction structure. The three selectional restriction structures at the psychological, expressive and descriptive levels correspond to and reflect each other, as if they were building a house. The selectional restriction elements go through continuous changes during its course of the three levels but their correspondent relations on the three levels seem quite clear: in the psychological level, some originate from concepts through the memory channel, some usages about the scale of selectional restriction objects through the linguistic consciousness; on expressive level, some come from words and concepts and some from lexical gap, like the saturated selectional restriction concept of qiāo(敲, "knock"); at the descriptive level, some stem from sememes and meaning clusters. Their corresponding relations are shown in the following However, the seemingly explicit hierarchical structure cannot be justified by the following aspects, turning to a strange loop.
Firstly, categories available for mastering the saturated selectional restriction elements can only be concepts or words on psychological and expressive levels and the meaning cluster on descriptive level. In the beginning it is repetitive work to list another equivalent sememe when a concept exists, for the latter takes an advantage over word in that a word cannot cover all synonymous variant under the same concept (e.g. the word gāoxì ng(高兴, happiness) does not cover the meaning of xǐyuè(喜悦, "joy"), yúkuà i (愉快, "elation"), huānlè(欢乐, "rapture")etc.) while word takes an advantage over concept in its real existence in language. Besides, in such cases there is no available concept and only substitute phrase in expression, meaning cluster can glue the meaning to be expressed with its ability to capture and glue the unintentionally-expressed and non-emerging content for comparison and reflection. Therefore, it is best to start from the concept and word on psychological and expressive level and the meaning cluster on descriptive level but they cannot concur in any set of the three selectional restriction structures, which leads to the level-crossing of the three but they are complementary.
Secondly, when the same selectional restriction structure is located at the three different levels of psychology, expression and description, the existing or implicit structural linkage, namely selectional restriction relation between the given word and the selectional restriction elements is invariant (e.g. the three levels in 34-36 are all "process-instrument" relation). However, the strange loop linked up among the three levels according to abstract degree gets this invariant selectional restriction linkage involved. On psychological and expressive level, the given word and selectional restriction word / concept can finally become the replicate of the contents of selectional restriction relations if they keep on climbing the abstract ladder simultaneously (see a and b in 34 and 35). Meanwhile, on descriptive level, they are also the replicate of certain core sememes of the original given word and the original selectional restriction word/concept (hyponymically related to the rational meaning, like the core sememe for hóuzi(猴子, "monkey") being [+animal](Ma Qinghua, 2000:182). See the selectional restriction unit and linkage in 34b and 35b and the core sememe in 36. Descriptive level provides explanation for the expressive level and description can only be carried out with the abstraction of the higher level so the descriptive level tries its best to be more abstract. For this reason, mostly the superordinate of a word must be a sememe on its descriptive level and the descriptive level mainly takes saturated structure as described object. In the expressive level, it is conducted for the most part in the lowest position of the ladder of the abstraction due to the rarer presence of saturated structure (usually considered information redundant and bad structure) and over-saturated structure (formed for ambiguous expression) than the non-saturated structure. A given word can also present itself at the expressive level of the saturated selectional restriction structure with the same identity, as in 37a but cannot at the descriptive level of its selectional restriction structure except for its substitution by its superordinate as in 37b.
3.4.1 The intertwine between the metalanguage and the object language on the descriptive level The intertwine between the metalanguage and the object language on the descriptive level widely spreads in the relation between language and its definition, which has been a common sense of linguistics. (Ma Qinghua, 2000:26) In 38a the object language is included in the metalanguage which form an intertwined nonsense structure. (Ma Qinghua, 2006: 372-373) However, if we make slight changes or make some efforts, by means of synonymous substitution of gǎndòng (感动, "move") for dǎdòng (打动, "touch") ① though substitution of synonym causes increasingly greater errors, the descriptive methods can be accepted, the reason for which may be that it acquires its own value and meaning during the equivalent substitution. "Structure by synonym substitution gains meaning", (D. R. Hofstadter, 2001:66-67) which means that describing the same thing in different ways generates a number of expressions which can gain acceptable meaning by synonymous substitution.
(38) a.dǎdòng(打动, "touch"): [♀/+touchable]; b. dǎdòng(打动, "move"): [♀/+movable] 3.4.2 The intertwinement at the level of syntax and lexical collocation The nonsense caused by internal intertwinement at the same level can be removed through resort to the intertwinement between the level of syntax and lexical collocation. For example, the intertwined structure between metalanguage and object language presents itself in the lexical intertwinement between different syntactical levels at the expressive level, as in 39, the real expression of 38.
The above analysis indicates that intertwinement feature is an important basis on which lexical selectional restriction linkage exist and function and also a significant part of the lexical selectional restriction theory. As some scholars say, the strange loop of hierarchical intertwine is no longer of negative value and is of positive value or harmony at higher level. (D. R. Hofstadter, 2001)

Conclusion
This paper gives a description and analysis of the internal determinacy of selectional restriction structure, the determinate linkage between the selectional restriction structure and the intertwinement features of selectional restriction structure. It is necessary for second-language vocabulary teachers to attach great importance to the internal determinacy of lexical selectional restriction structure and determinate relation between selectional restriction structures. Since intertwinement feature is the basis of selectional restriction linkage, intertwinement theory enables the scientific analysis of selectional restriction structure and the whole lexical system and guarantees their application to practice. In additions, the process of argumentation for the intertwinement feature of selectional restriction structure enables us to recognize the intertwinement nature of the whole linguistic system.