The Interplay of Synonymy and Polysemy

§4.1, §4.2.1, §4.3.1 and §4.3.2. 
This study explores the features individually. That is, each variable is measured one at 
a time. Tummers et al. (2005:243) argue that analyzing a series of features independently of 
each other is a simplification of the language. In fact, language is complex and parameters are 
capable of interacting. But Liu (2010:62) points out that the way that near synonyms differ 
will often vary depending on the words under study. This study is in fact the first step of 
exploring which features could be relevant to these four 
throw
-verbs. Gilquin (2010) begins 
her analysis of periphrastic causative constructions by exploring each feature individually 
using chi-square tests, and only afterwards moves to statistical measures (linear regression) to 
study all features concurrently. There is a value, then, in exploring features individually, if 
only as a first step.

---

39 | 
A second step would be to follow Gries (2006), Divjak (2006) Divjak & Gries (2008), 
Gries & Otani (2010) and Gilquin (2010). They use statistical tools that take into account all 
features simultaneously. In Divjak & Gries (2008:192-193) this results in a dendogram 
represents the different degrees of semantic similarity between the verbs in their study. The 
verbs that are clustered or connected earlier are semantically closer than the verbs that are 
connected later. A statistical study analyzing all kinds of semantic, syntactic and 
morphological features simultaneously would be a further step in the study of these verbs.
3.2 Collostructional Analysis
 
The 400-sentence corpus is the major source of data. Through manual annotation and 
semantic analysis, we can paint a picture of the data collected. The statistical tests ensure that 
the patterns are significant and representative of the verbs under study. Still, it is of interest to 
be able to explore a larger data set. The benefits of analyzing a larger set are two-fold. First, it 
allows us to confirm some of the results of the 400-corpus. Secondly, it allows us to discover 
patterns that happened to not be represented in the sample corpus.
A second limited corpus study was performed, again using the 1900s subcorpus of the 
CdE. The data for the collostructional analysis was extracted in the months of March and 
April 2011. The purpose of this second study was to explore the collocational patterns of 
these verbs. Many studies have used collocational patterns in studying verbs. Dąbrowska 
(2009) used collocational patterns to show that speakers can distinguish between near-
synonymous motion verbs. In her study, collocates include different kinds of words that 
occurred with each verb within a sentence, such as nouns, adverbs and prepositional phrases. 
Bybee & Eddington (2006) study four Spanish verbs that often occur in verbal periphrasis. 
Their relevant data includes only the adjective that occurs as the second element in the 
periphrasis. In both cases, the collocational patterns revealed that the verbs have specific 
preferences for the words they co-occur with.
These two studies implement two different definitions of collocations; one including 
all types of words in all types of relations with the verb and another where collocates are 
limited to specific types of words in specific relations to verb. Stefanowitsch & Gries (2003) 
argue that this second operationalization of collocations is the most insightful. They propose 
a method for analyzing collocational preferences which they term