Joint Transition-Based Models for Morpho-Syntactic Parsing: Parsing Strategies for MRLs and a Case Study from Modern Hebrew

Abstract

In standard NLP pipelines, morphological analysis and disambiguation (MA&D) precedes syntactic and semantic downstream tasks. However, for languages with complex and ambiguous word-internal structure, known as morphologically rich languages (MRLs), it has been hypothesized that syntactic context may be crucial for accurate MA&D, and vice versa. In this work we empirically confirm this hypothesis for Modern Hebrew, an MRL with complex morphology and severe word-level ambiguity, in a novel transition-based framework. Specifically, we propose a joint morphosyntactic transition-based framework which formally unifies two distinct transition systems, morphological and syntactic, into a single transition-based system with joint training and joint inference. We empirically show that MA&D results obtained in the joint settings outperform MA&D results obtained by the respective standalone components, and that end-to-end parsing results obtained by our joint system present a new state of the art for Hebrew dependency parsing.

Crime Linkage Based on Textual Hebrew Police Reports Utilizing Behavioral Patterns

“…Utilizing term frequency-inverse document frequency, we represented each of the police reports with its own sparse vector. In order to improve the accuracy of the TF-IDF representation with Hebrew, we performed the following steps: (1) we removed Hebrew stop words, and (2) we used Hebrew tokenization and lemmatization utilizing the 'YAP' parser tool [25].…”

“…In our experiments, the police reports used are written in Hebrew, which makes the extraction task very difficult [17,25,37]. Hebrew is a morphologically rich language, and its text tends to be very ambiguous; each token can have a large set of possible inflections, according to quantity, gender, and tense [19].…”

A Graph-Based Framework for Structured Prediction Tasks in Sanskrit

“…Assuming the inflection to be in nominative-case, this information enables the nominal to form one of the two possible syntactic relations with the main verb in the sentence, namely, kartā (subject) or karma (object) 2 , in the syntactic analysis of the sentence. The cyclic dependency between morphological and syntax-level tasks is well known (Tsarfaty 2006), and these tasks are often solved jointly (More et al 2019). Similarly, the potential error propagation from word segmentation to its downstream tasks in pipeline models is also well established for multiple languages (Hatori et al 2012;Zhang and Yang 2018).…”

“…The performance of a system depends highly on the choice of feature function used for the task. In MRLs, hand-crafted features still form a crucial component in contributing to the performance of the state-of-the-art systems for tasks such as morphological parsing and dependency parsing More et al, 2019;Seeker and Ç etinoglu 2015). But Krishna et al (2018) learn a feature function using the Path Ranking Algorithm (PRA) (Lao and Cohen 2010) for the joint task of word segmentation and morphological parsing.…”

“…The tasks in the hierarchy are collectively termed as morphosyntactic tasks as each of these tasks either require morphosyntactic information in input or is an upstream task to such tasks (e.g., word segmentation). The joint modeling of such related tasks in a hierarchy has empirically been shown to perform better than a pipeline-based approach (Seeker and Ç etinoglu 2015;More et al 2019). We jointly model the morphological parsing (MP), and word segmentation (WS) tasks, similar to Krishna et al (2018).…”