[Paper Review] Building an Ellipsis-aware Chinese Dependency Treebank for Web Text

Xuancheng Ren, Xu Sun+4 more|arXiv (Cornell University)|Jan 20, 2018

Natural Language Processing Techniques3 citations

TL;DR

This paper introduces the first Chinese dependency treebank specifically designed to handle ellipsis in web text, focusing on Weibo microblogs. By restoring omitted elements and preserving contextual information in 572 sentences across 319 Weibo posts, the authors create a resource to improve dependency parsing for informal, spoken-like language, addressing a key challenge in NLP for user-generated content.

ABSTRACT

Web 2.0 has brought with it numerous user-produced data revealing one's thoughts, experiences, and knowledge, which are a great source for many tasks, such as information extraction, and knowledge base construction. However, the colloquial nature of the texts poses new challenges for current natural language processing techniques, which are more adapt to the formal form of the language. Ellipsis is a common linguistic phenomenon that some words are left out as they are understood from the context, especially in oral utterance, hindering the improvement of dependency parsing, which is of great importance for tasks relied on the meaning of the sentence. In order to promote research in this area, we are releasing a Chinese dependency treebank of 319 weibos, containing 572 sentences with omissions restored and contexts reserved.

Motivation & Objective

• Address the challenge of ellipsis in informal, user-generated web text, particularly in Chinese microblogs.
• Improve dependency parsing performance by explicitly modeling omitted elements rather than treating them as structural gaps or orphans.
• Create a high-quality, context-aware treebank that supports research in syntactic and semantic analysis of colloquial Chinese.
• Provide a resource that distinguishes ellipsis from pronoun-dropping and empty categories, enabling more accurate linguistic modeling.
• Facilitate the development of robust parsing frameworks for Web 2.0 text by offering a dataset with restored syntactic structure.

Proposed method

• Collected 319 Weibo messages containing ellipsis from real user posts.
• Restored omitted words (e.g., verbs, noun phrases) based on contextual inference, ensuring syntactic and semantic plausibility.
• Preserved the original context around each sentence to maintain discourse-level coherence.
• Annotated the restored sentences using dependency parsing with a focus on accurate syntactic relations.
• Designed annotation guidelines to ensure consistency in identifying and restoring ellipsis across diverse, informal constructions.
• Released the dataset publicly via GitHub to support reproducibility and further research in dependency parsing and ellipsis resolution.

Experimental results

Research questions

• RQ1 How can ellipsis in informal Chinese web text be systematically identified and restored to improve dependency parsing?
• RQ2 What are the most frequent types of ellipsis in Chinese microblogs, and how do they affect syntactic structure?
• RQ3 To what extent does restoring omitted elements improve parsing accuracy compared to treating ellipsis as a gap or orphan?
• RQ4 How does context influence the correct restoration of ellipsis in spoken-like, colloquial Chinese?
• RQ5 Can a dedicated treebank with restored ellipsis enhance the performance of neural dependency parsers on informal text?

Key findings

• The authors successfully constructed a Chinese dependency treebank of 572 sentences from 319 Weibo posts with ellipsis restored and context preserved.
• The dataset is the first of its kind to focus specifically on ellipsis in Chinese web text, distinguishing it from general treebanks or those treating ellipsis as empty categories.
• Restoring omitted elements enables more accurate syntactic and semantic parsing, especially in spoken-like, informal language where ellipsis is pervasive.
• The treebank provides a foundation for training and evaluating dependency parsers that can handle real-world, colloquial linguistic phenomena.
• The resource is publicly available at https://github.com/lancopku/Chinese-Dependency-Treebank-with-Ellipsis, enabling broader research in NLP for informal text.
• The study highlights the limitations of existing approaches like Universal Dependencies, which often treat ellipsis via orphan relations or dependency promotion, leading to confusing or isolated structures.