Cross-Lingual Annotation Projection for Weakly-Supervised Relation Extraction

Abstract

Although researchers have conducted extensive studies on relation extraction in the last decade, statistical systems based on supervised learning are still limited, because they require large amounts of training data to achieve high performance level. In this article, we propose cross-lingual annotation projection methods that leverage parallel corpora to build a relation extraction system for a resource-poor language without significant annotation efforts. To make our method more reliable, we introduce two types of projection approaches with noise reduction strategies. We demonstrate the merit of our method using a Korean relation extraction system trained on projected examples from an English-Korean parallel corpus. Experiments show the feasibility of our approaches through comparison to other systems based on monolingual resources.

Distributed Word Representation Learning for Cross-Lingual Dependency Parsing

“…For cross-lingual annotation projection methods, both the word alignment training step and the annotation projection step can introduce errors or noise. Thus much work developed in the literature has focused on designing robust projection algorithms such as graph-based projection with label propagations (Das and Petrov, 2011), improving projection performance by using auxiliary resources such as Wikipedia metadata (Kim and Lee, 2012) or WordNet (Khapra et al, 2010), or boosting projection performance by heuristically modifying or correcting the projected annotations (Hwa et al, 2005;Kim et al, 2010). Some work has also proposed to project the discrete dependency arc instances instead of treebank as the training set .…”

Name Entity Recognition for Malay Texts Using Cross-Lingual Annotation Projection Approach

“…The scope of this research is focusing on projecting the annotation information from the SL (English) to the TL (Malay) within a non-pre-aligned bilingual corpora or parallel corpus. This task is also referred as cross-language annotation projection [18] and tools through the annotation projection process. The challenge is to develop an algorithm to automatically align the sentences from two languages of different structures and morphology, a rich-resourced language (English) and a less-resourced language (Malay) and thus to project the English annotations across the parallel corpus with higher accuracy.…”

“…This can be achieved using some existing tools which will be discussed in detail in the next sub topics. Annotation projection using pre-aligned parallel corpus is demonstrated successfully in projecting coreference resolution in English-Portuguese parallel corpus [26], relation detection in English-Korean parallel corpus [18], dependency analysis in English-Swahili parallel corpus [27], semantic roles in English-German parallel corpus [28] and syntactic relations in English-Romanian parallel corpus [29].…”

Translation approaches in Cross Language Information Retrieval

“…Seokhwan Kim et.al [18] proposed that in weakly-supervised relation a new external resource is leverage parallel corpora. This method utilizes parallel corpora by presenting the annotations in source language which is generated by relation extraction system to acquire training examples in target language.…”