From Abstract Syntax to Universal Dependencies

LiLT volume 13, issue 3 October 2016 From Abstract Syntax to Universal Dependencies Prasanth Kolachina and Aarne Ranta, University of Gothenburg Abstract. Abstract syntax is a semantic tree representation that lies between parse trees and logical forms. It abstracts away from word order and lexical items, but contains enough information to generate both surface strings and logical forms. Abstract syntax is commonly used in com- pilers as an intermediate between source and target languages. Gram- matical Framework (GF) is a grammar formalism that generalizes the idea to natural languages, to capture cross-lingual generalizations and perform interlingual translation. As one of the main results, the GF Resource Grammar Library (GF-RGL) has implemented a shared ab- stract syntax for over 30 languages. Each language has its own set of concrete syntax rules (morphology and syntax), by which it can be generated from the abstract syntax and parsed into it. This paper presents a conversion method from abstract syntax trees to dependency trees. The method is applied for converting GF-RGL trees to Universal Dependencies (UD), which uses a common set of la- bels for different languages. The correspondence between GF-RGL and UD turns out to be good, and the relatively few discrepancies give rise to interesting questions about universality. The conversion also has po- tential for practical applications: (1) it makes the GF parser usable as a rule-based dependency parser; (2) it enables bootstrapping UD tree- banks from GF treebanks; (3) it defines formal criteria to assess the informal annotation schemes of UD; (4) it gives a method to check the consistency of manually annotated UD trees with respect to the anno- tation schemes; (5) it makes information from UD treebanks available 1 2 / LiLT volume 13, issue 3 October 2016 for the construction and ranking of GF trees, which can improve GF applications such as machine translation. The conversion is tested and evaluated by bootstrapping two small treebanks for 31 languages, as well as comparing a GF version of the English Penn treebank with the UD version. 1 Introduction Computational syntax can work on different levels of abstraction. The lowest level normally used when processing written text is strings of to- kens ("words"). But it is often useful to work with more abstract struc- tures: part-of-speech (POS) tagged lemma sequences, phrase structure trees, dependency trees, or some kind of logical forms. Raising the level of abstraction often gives new ways to relate differ- ent languages to each other. Thus tagged lemmas enable the separation of surface strings from word senses, which can be useful, for instance, in factored machine translation. Logical forms ideally ignore all language- related features, and express just the pure propositional meaning. But what about syntax trees? Traditional phrase structure trees preserve surface words and constituent order and are hence language-dependent. Dependency trees are often a bit more abstract, treating the word order as irrelevant and lemmatizing the words. Sharing the POS tags and de- pendency labels between languages increases this potential to abstract over languages. Universal Dependencies …

Content truncated to 500 words