ausblenden:
Schlagwörter:
semantic tightness, corpora, typology, mutual information, alternations
Zusammenfassung:
One issue that has received a lot of attention in quantitative coprpus-based Construction Grammar has involved describing and enumerating the permitted slot fillers for different constructions (Stefanowitsch & Gries 2003; Levshina & Heylen 2014; Perek & Hilpert 2015, to name just a few). The purpose of the present paper is to add to the cross-linguistic dimension of this research, by focusing on the measure of association strength between alternating constructions and lexical slot fillers. This allows us to create a multidimensional constructional typology of thirty different languages.
It has been argued that some languages have more semantic restrictions on their syntactic roles (e.g. a transitive subject must be an agent and not an experiencer, instrument or locative) while others have less. For instance, English has fewer semantic restrictions on its grammatical subjects than German and Russian (Plank 1984). This means that English is ‘looser’, and German and Russian are ‘tighter’ in their form-to-meaning mappings (Hawkins 1986: 121–127, 1995; Müller-Gotama 1994). In previous work (Levshina 2020), we used corpora to measure the strength of association between nouns and the main syntactic roles they could fill, which served as a proxy for semantic tightness. This enabled us to quantify cross-linguistic differences in the strength of construction–lexeme associations, in accordance with the gradient approach of usage-based Construction Grammar. In this paper we extend this quantitative method to verbs, focusing on three alternations.
Large samples of text in thirty different languages were taken from the Leipzig Corpora Collection (Goldhahn et al. 2012). They cover online news content. The texts are morphologically and syntactically annotated with the help of the Universal Dependencies tools (Straka & Straková 2017). We extracted the following information:
- Frequencies of verbs in transitive and intransitive constructions that have the same lexical subject and object (e.g. The door opened vs. Someone opened the door). If a verb can be used in both constructions, it is known as P-lability (cf. Dixon 1994).
- Frequencies of verbs in transitive and intransitive constructions with the same lexical subject (e.g. The couple danced vs. The couple danced a waltz), also known as A-lability.
- Frequencies of verbs with the same lexical transitive subject and intransitive oblique (e.g. This tent sleeps six vs. Six can sleep in this tent).
The choice of these alternations was motivated by the availability of sufficient data in the corpora. Only active and non-reflexive forms of the verbs were considered. We also collected the frequencies of individual nouns as transitive subject, intransitive subject, object and oblique. All these frequencies were used to compute mutual information, which represents the strength of association between the lexemes and the constructional slots in each language.
Our initial results reveal some correlations between the measures. One of them is displayed in Figure 1, which shows the strength of association between nouns and their syntactic roles (horizontal axis) against the strength of association between verbs and constructions of the P-lability type (vertical axis). The correlation is obviously not perfect (r = 0.48, p = 0.007), but as the semantic restrictions on syntactic roles increase and become tighter, as seen in the higher mutual information scores horizontally from English on the left to Korean on the right, so P-lability also becomes more restricted and more indicative of the relevant (transitive or intransitive) construction type, as seen in the rising overall mutual information scores vertically. The noun – syntactic role associations tend to be strong in verb-final languages, which has been explained by the pressure to avoid costly reanalysis (Hawkins 1995) and by the greater efficiency of assigning more tightly constrained theta-roles early prior to the verb itself (Bornkessel 2002). The verb associations with constructions seem to be correlated with this, but also driven (at least, partly) by other factors, which require further investigation.
