Article Preview
TopIntroduction
Numerous research projects in linguistics, psychology and computer science deal with how languages manage to encode motion, i.e. the way languages encode changing the position of an entity in relation to another entity (Vulchanova & van der Zee, 2013, p. 2).
In this paper, we deal with the semantics of motion verbs with the prefix “do-” and the adlativity feature of the motion prefixed verbs in Croatian, while exploring the possibility for a cross-linguistic approach to semantic role labelling (more precisely the labelling of DIR3 (Direction-To) and Goal semantic roles in different languages), as well as solving semantic ambiguities in machine translation post-editing or post-processing.
The paper aims to present how semantic role labeling of adlativity (expressing motion and direction) could help in computational linguistics for further text processing and in practical implementation, as in machine translation where morphological and syntactic errors influence the perceived quality of machine-translated text (Seljan et al., 2015).
Semantic role labeling is a natural processing task that aims to reveal who did what to whom, when and why.
From the practical viewpoint, semantic role labelling is valuable for a whole range of natural processing tasks, such as question answering (Yih et al., 2016; Fitzgerald, He, & Zettlemoyer, 2018), event detection (Mikelic Preradovic et al, 2013), machine reading comprehension, textual entailment and context-oriented dialogue (Zhang et al., 2018), since deep learning models do not really understand the natural language texts (Mudrakarta et al., 2018). Also, semantic role labelling proved to be useful for machine translation (Shi et al., 2016) and discourse relation sense classification (Mihaylov & Frank, 2016).
The main contribution of this paper is to present the framework of semantic role labelling for adlative meaning in 7 languages, as presented in Table 2.
Motion encoding in the spatial language (the part of language describing perceived space) refers to how languages encode path of motion (i.e. direction) which has a beginning point (source of motion), a middle point (trajectory) and an endpoint or goal (Luraghi, Nikitina & Zanchi, 2017, p. 2). Directional meaning can be categorized as follows: (1) the adlative that expresses direction “moving closer to”; (2) the ablative that expresses direction “moving away from” and (3) the perlative that expresses motion which is directionally unspecified and represents only the path of the moving object (Mikelić Preradović, Boras & Lauc, 2013). Talmy (2000) differentiates satellite-framed languages (S-languages) where the path is lexicalized as a “satellite” to the verb (e.g. English, Slavic and Germanic languages) and verb-framed languages (V-languages) where the path is lexicalized in the root of the motion verb, while the meaning of a verb is provided by its prefix (e.g. Romance languages). The satellite may be either a verb prefix (like Russian or Croatian inseparable verb prefixes or German separable and inseparable verb prefixes) or a free word (like English verb particles). Regarding differences in verb prefixation of the motion verb, Dickey (2010) distinguishes the North Slavic allative prefixation (e.g. Russian and Czech) and South Slavic allative prefixation (e.g. Croatian and Serbian), where the first group of languages generalizes “pri-” (“to”) as the prefix of the motion verbs that signal the crossing of the boundary of the goal, while the latter generalizes “do-” (“to”) as the prefix of the motion verbs that do not assert the crossing of the boundary of the goal, but only the traversal of a trajectory up to the goal. In other words, apart from the difference between “pri-” and “do-”, goal prepositional phrases in the North Slavic languages specify the crossing of the boundary, while the goal prepositional phrases in the South Slavic languages can specify the boundary crossing or just reaching the boundary.