Diachronic proximity vs. data sparsity in cross-lingual parser projection. A case study on Germanic

Abstract

For the study of historical language varieties, the sparsity of training data imposes immense problems on syntactic annotation and the development of NLP tools that automatize the process. In this paper, we explore strategies to compensate the lack of training data by including data from related varieties in a series of annotation projection experiments from English to four old Germanic languages: On dependency syntax projected from English to one or multiple language(s), we train a fragment-aware parser trained and apply it to the target language. For parser training, we consider small datasets from the target language as a baseline, and compare it with models trained on larger datasets from multiple varieties with different degrees of relatedness, thereby balancing sparsity and diachronic proximity.Our experiments show (a) that including related language data to training data in the target language can improve parsing performance, (b) that a parser trained on data from two related languages (and none from the target language)can reach a performance that is statistically not significantly worse than that of a parser trained on the projections to the target language, and (c) that both conclusions holds only among the three most closely related languages under consideration, but not necessarily the fourth.The experiments motivate the compilation of a larger parallel corpus of historical Germanic varieties as a basis for subsequent studies.

Machine Translation and Automated Analysis of the Sumerian Language

“…Other difficulties arise in the annotation projection and machine translation from the fact that the Sumerian language does not have any modern descendants. This is particularly important for the annotation projection, as previous studies have shown that diachronic relatedness is an important factor that affects the quality of annotation projection (Sukhareva and Chiarcos, 2014). Thus, we plan to conduct our pilot experiments on modern languages such as Turkish and Basque that are grammatically similar to Sumerian (agglutinative, split-ergative) to guarantee the scalability of our implementation, but more importantly to be able to conduct experiments in parallel with the morphological and syntactic annotation of the Sumerian texts.…”

If all you have is a bit of the Bible: Learning POS taggers for truly low-resource languages

“…It has also been used in cross-lingual POS tagging Fossum and Abney, 2005), NP-chunking ) and cross-lingual dependency parsing (Sukhareva and Chiarcos, 2014) before. and Fossum and Abney (2005) use word-aligned parallel translations of the Bible to project the predictions of POS taggers for several language pairs, including English, German, and Spanish to Czech and French.…”

Distantly Supervised POS Tagging of Low-Resource Languages under Extreme Data Sparsity: The Case of Hittite

“…Performance of distantly supervised methods such as annotation projection or cross-lingual tool adaptation depends on the diachronic relatedness between the source and the target languages. For example, annotation projection from modern English into middle English gives better results than into old English because middle English grammatically and lexically resembles modern English much more than Old English (Sukhareva and Chiarcos, 2014). Annotation projection is thus typically applied to related languages (Tiedemann and Agic, 2016).…”