The Interplay of Synonymy and Polysemy

Graph 1. 
Occurrence of the four 
throw
-verbs in the CdE 
1900s subsection.
 
0
500
1000
1500
2000
2500
3000
arrojar
echar
lanzar
tirar
Nonfinite
Finite

---

35 | 
counts could be explored using statistical measures. Chi-square tests (explained in the 
following paragraphs) become unreliable with very low counts; any value below 5 is not 
safely tested by chi-square. Limiting the amount of variables per parameter helps to keep the 
counts higher.
Sentences from the corpora also contain phrases with other functions, such as phrases 
that detail the reason that an object was thrown, the manner in which it was thrown and the 
time when it happened, for example. The 
INITIATOR
,
MOVANT 
and 
DIRECTIONAL 
are the most 
frequent in the data set. These phrases are in a semantic and syntactic relation to the verb and 
could be argued to be core elements of the meaning of each verb (Morimoto 2001, ADESSE, 
Spanish FrameNet).
Previous studies have explored (the semantic make-up of) the lexical items that 
appear in specific (syntactic) relations to the verb or noun under study. Arppe & Järvikivi 
(2007) explored the types of subjects that occurred with synonymous Finnish verbs. They 
found that each verb had different preferences on the types of subject. They used 
morphological features in some cases, but the focus was on the semantics of the subject. Liu 
(2010) was able to organize synonymous adjectives into a scale of meaning by first exploring 
the kinds of nouns each modified.
Bybee & Eddington (2006) analyzed Spanish verbs of becoming focusing primarily 
on the types of words that combined with these verbs. Then they organized these words into 
several groups based on semantic similarity. This categorization showed that the verbs 
typically combine with specific words with high frequency. The less commonly occurring 
nouns tended to appear with more than one of the verbs. Wulff (2006) studied the verbs that 
occur in the go-V and go-and-V constructions. She argues that the different types of verbs 
that appear with each construction signal differences in meaning between the constructions. 
Meaning and syntactic information can be explored in conjunction. Divjak (2006), 
Divjak & Gries (2008) and Gries & Otani (2010) propose using multiple levels of linguistic 
features instead of focusing on a single feature. They annotate their corpora with ID tags 
which include morphological, syntactic, collocational and semantic information. They argue 
that the conjunction and interplay of the various features give a more complete picture of the 
closeness of synonyms (and antonyms). Divjak & Gries (2008) use statistical analyses of 
these ID tags to derive clusters of verbs, showing some to be closer in meaning than others.
Divjak (2006) argues that the distribution of synonymous verbs in different 
constructions and the distribution of collocates that appear in the slots of those constructions 
can be used to measure the similarity of verbs. This is the focus chosen for this study. The 

---

36 | 
focus is on the participants or arguments that appear with each verb. Exploring the participant 
roles gives us data on the common collocates for each verb, and also gives us indications of 
the types of constructions that each verb can participate in.