Collostructional analysis is grounded in two frameworks—one theoretical and one methodological (Gries & Stefanowitsch, 2004a). First of all, the theoretical framework is based on the Construction Grammar, which views various levels of formal abstraction (i.e. morphemes, words, idioms, etc.) as a meaningful linguistic unit in its own right. The two important
assumptions in Construction Grammar, introduced in Section 1.2, are repeated here: (1) if a word
can occur in a construction, then the word must be semantically compatible with the meaning of the construction (Goldberg, 1995). (2) if two constructions are syntactically distinct, then they must be semantically or pragmatically distinct (Goldberg, 1995). Therefore, collostructional analysis emerges as a useful method to study the relationship between words and the
grammatical structures they cooccur with (Stefanowitsch & Gries, 2003).
The methodological framework is based on the quantitative corpus linguistics, featuring three methodological characteristics. First, naturally occurring linguistic data from a
representative and balanced corpus is always needed for the linguistic investigation with this approach. Second, the linguistic phenomena in question should be extracted exhaustively from the corpus. Third, the data should be submitted to strict quantification and inferential statistics to determine the importance of the relationships. The Collostructional analysis highlights the importance of measuring the association strength between a construction and a word. It is quite different from the simple frequency-count method, since collostructional analysis builds a model of probability and can be used for generalization. (Gries & Stefanowitsch, 2004a, 2004b;
Stefanowitsch, 2013; Stefanowitsch & Gries, 2003).
There are three major variant techniques of collostructional analysis: the simple collexeme analysis, distinctive collexeme analysis and covarying collexeme analysis. The multiple collexeme analysis, which is the technique used in the present study, is an extension of the distinctive collexeme analysis. The three techniques in the collostructional analysis are briefly summarized in (6).
(6)
a. Simple collexeme analysis: the analysis that studies co-occurrence relation between a lexical construction (i.e. a word) and a grammatical construction, e.g. the V slot in the [X think nothing of Vgerund] (Stefanowitsch & Gries, 2003)
b. Distinctive collexeme analysis: the analysis that studies one slot in two semantically or functionally similar constructions, e.g. the different semantics of verbs in the
ditransitive and to-dative constructions of (Gries & Stefanowitsch, 2004b)
b’ Multiple Distinctive collexeme analysis: the extended analysis that studies one slot in more than two semantically or functionally similar constructions, e.g. the V-slot in the causative construction (Gilquin, 2006).
c. Covarying collexeme analysis: the analysis that studies the interaction between two slots within a construction, e.g. V1 and V2 in the [X V1 into V2gerund] causative
construction (Gries & Stefanowitsch, 2004a)
Simple collexeme typically studies co-occurrence relation between a lexical construction (i.e. a word) and a grammatical construction. For example, Stefanowitsch and Gries (2003) investigate the V slot in the [X think nothing of V]. The meaning of the construction refers to ‘to have low opinion of something’. The study adopts a simple collexeme analysis to uncover which verbs are more strongly attracted to the particular slot of this construction. Then those
collexemes are ordered according to their association strength to the construction. In terms of interpreting the result, the collostructional strength refers to the association strength between the construction and the verb and is represented by p-value (how the p value is calculated will be demonstrated in Chapter 3). The smaller the p-value, the stronger the attraction. The ranking
does pick out verbs denoting possible risk like mortgage, confine, motor, leap and fly, and verbs denoting an undesirable situation like haggle, rip, abandon. As a result, the simple collexeme analysis provides a more statistics-based result than purely intuitive analysis by showing the degree of attraction to a particular construction.
The covarying collexeme analysis investigates the interaction between two slots within a construction. Two terms that are attracted to the slots can be further analyzed to uncover the semantic coherent patterns of the constructions using exploratory techniques, such as cluster analysis. For example, Gries and Stefanowitsch (2004a) look into the interaction between V1 and V2 slot in the [X V1 into V2gerund] causative construction. Most of the previous studies only discussed about the semantics of the verbs appearring in V1 slot, which often fall into categories like TRICKERY (i.e. trick, fool), PHYSICAL FORCE (i.e. force, bully), and VERBAL
PERSUASION (i.e. coax, goad) (Stefanowitsch & Gries, 2003). The study aims to explore whether different semantic types of V1 might prefer different types of verbs in the V2 slot. They proposed a covarying collexeme analysis for identifying pairs of verbs that are strongly
associated with each other in that construction, and the results support the assumption that different kinds of verbs in V1 slot would attract different kinds of verbs in V2 slot. Take the second significant covarying collexeme (i.e. torture and confess) for example. Both torture and confess are semantically related because they are both related to an act of communication that is not out of the speaker’s willingness. By using the covarying collexeme technique, we are able to understand the possible interaction between two potential collocates within a construction.
Distinctive collexeme analysis studies one slot in two or more semantically or
functionally similar constructions. For example, Gries and Stefanowitsch (2004b) use the method to investigate the dative alternation in English. They look into the ditransitive construction like
John sent Mary the book and the to-dative construction like John sent the book to Mary. With the aid of distinctive collexeme analysis, we learn that in ditransitive construction, give is the verb with the strongest association strength to the construction. The verb encodes a face to face interaction between the agent and the recipient. Similar examples found in the distinctive
collexemes are tell, show, offer, teach, ask. On the other hand, in to-dative construction, the verb with strongest attraction to the construction is bring, which indicates some distance between the agent and the recipient and the agent must take some time to complete the action. Similar examples found in the distinctive collexemes are take, pass, supply. To sum up, the distinctive collexeme analysis helps us to identify the possible difference between two or more semantically similar construction from a corpus-driven quantitative perspective.
The multiple collexeme is an extension of the distinctive collexemes analysis, where more than two constructions are under investigation. Gilquin (2006) adopts this technique to investigate the 10 causative constructions as shown in Table 2-5. She pointed out that little is done on the lexical preference with the respect to V slot in these causative constructions. The result from multiple distinctive collexeme analysis indicates that constructions show very strong preference for different groups of verbs in the non-finite verb slot. [X MAKE Y Vinf]
construction attracts verbs that cause a process that is not directly dependent on the CAUSEE. In other words, they are descriptive verbs like look, appear, seem and verbs denoting mental
processes or psychological states like think and wonder without too much volition involved.
The multiple distinctive collexeme analysis again shows its power in determining the subtle yet important differences among near-synonymous constructions. The present study will also adopt the same analysis method to identity the semantic difference among the three spatial particle constructions in Chinese.
Table 2-5 The Causative Constructions And The Examples
Construction Example
[X CAUSE Y Vto-inf] cause the price to fall
[X GET Y Vto-inf] get Jes to do it
[X GET Y Vpp] get everything sorted out
[X GET Y Vprp] get these earphones working
[X HAVE Y Vinf] have Roche inspect the hut
[X HAVE Y Vpp] have the blade sharpen
[X HAVE Y Vprp] have a blow lamp going
[X MAKE Y Vinf] make our home seem so small
[X BE made Vto-inf] be made to pay
[X MAKE Y Vpp] make their voice heard
3 Methodology
This chapter illustrates the procedures of our data analysis. Section 3.1 gives a brief
overview of the corpus adopted in this study. Section 3.2 introduces the steps of data extraction.
Section 3.3 presents two analyses (the multiple collexeme analysis, and post-hoc semantic analysis) to the extracted data. Section 3.4 is the summary of the whole chapter.