Better Arabic Parsing: Baselines, Evaluations, and Analysis

Spence Green, Christopher D. Manning

2010 · 176 citations · 35 references

Concepts

TL;DR

The study investigates why Arabic constituency parsing underperforms, pinpointing syntactic ambiguity, annotation choices, and model design, and proposes a human‑interpretable grammar competitive with a latent‑variable PCFG while outlining how to build better models for three parsers. The authors identify understudied syntactic ambiguities, develop a human‑interpretable grammar that rivals a latent‑variable PCFG, and demonstrate how to construct improved models for three different parsers. They find that the Penn Arabic Treebank suffers from annotation inconsistencies despite comparable statistics, and that omitting gold segmentation in application settings lowers parsing F1 by 2–5 %.

Abstract

In this paper, we offer broad insight into the underperformance of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design. First, we identify sources of syntactic ambiguity understudied in the existing parsing literature. Second, we show that although the Penn Arabic Treebank is similar to other tree-banks in gross statistical terms, annotation consistency remains problematic. Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG. Fourth, we show how to build better models for three different parsers. Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2--5% F1.

References

35