GlyphCRM: Bidirectional Encoder Representation for Chinese Character with its Glyph

• URL: http://arxiv.org/abs/2107.00395v1
• Date: Thu, 1 Jul 2021 12:14:05 GMT
• Title: GlyphCRM: Bidirectional Encoder Representation for Chinese Character with its Glyph
• Authors: Yunxin Li, Yu Zhao, Baotian Hu, Qingcai Chen, Yang Xiang, Xiaolong Wang, Yuxin Ding, Lin Ma
• Abstract summary: Previous works indicate that the glyph of Chinese characters contains rich semantic information. We propose a Chinese pre-trained representation model named as Glyph CRM. It abandons the ID-based character embedding method yet solely based on sequential character images.
• Score: 31.723483415041347
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Previous works indicate that the glyph of Chinese characters contains rich semantic information and has the potential to enhance the representation of Chinese characters. The typical method to utilize the glyph features is by incorporating them into the character embedding space. Inspired by previous methods, we innovatively propose a Chinese pre-trained representation model named as GlyphCRM, which abandons the ID-based character embedding method yet solely based on sequential character images. We render each character into a binary grayscale image and design two-channel position feature maps for it. Formally, we first design a two-layer residual convolutional neural network, namely HanGlyph to generate the initial glyph representation of Chinese characters, and subsequently adopt multiple bidirectional encoder Transformer blocks as the superstructure to capture the context-sensitive information. Meanwhile, we feed the glyph features extracted from each layer of the HanGlyph module into the underlying Transformer blocks by skip-connection method to fully exploit the glyph features of Chinese characters. As the HanGlyph module can obtain a sufficient glyph representation of any Chinese character, the long-standing out-of-vocabulary problem could be effectively solved. Extensive experimental results indicate that GlyphCRM substantially outperforms the previous BERT-based state-of-the-art model on 9 fine-tuning tasks, and it has strong transferability and generalization on specialized fields and low-resource tasks. We hope this work could spark further research beyond the realms of well-established representation of Chinese texts.

Related papers

• DeepCalliFont: Few-shot Chinese Calligraphy Font Synthesis by Integrating Dual-modality Generative Models [20.76773399161289]
  Few-shot font generation, especially for Chinese calligraphy fonts, is a challenging and ongoing problem. We propose a novel model, DeepCalliFont, for few-shot Chinese calligraphy font synthesis by integrating dual-modality generative models.
  arXiv  Detail & Related papers  (2023-12-16T04:23:12Z)
• Chinese Text Recognition with A Pre-Trained CLIP-Like Model Through Image-IDS Aligning [61.34060587461462]
  We propose a two-stage framework for Chinese Text Recognition (CTR) We pre-train a CLIP-like model through aligning printed character images and Ideographic Description Sequences (IDS) This pre-training stage simulates humans recognizing Chinese characters and obtains the canonical representation of each character. The learned representations are employed to supervise the CTR model, such that traditional single-character recognition can be improved to text-line recognition.
  arXiv  Detail & Related papers  (2023-09-03T05:33:16Z)
• VQ-Font: Few-Shot Font Generation with Structure-Aware Enhancement and Quantization [52.870638830417]
  We propose a VQGAN-based framework (i.e., VQ-Font) to enhance glyph fidelity through token prior refinement and structure-aware enhancement. Specifically, we pre-train a VQGAN to encapsulate font token prior within a codebook. Subsequently, VQ-Font refines the synthesized glyphs with the codebook to eliminate the domain gap between synthesized and real-world strokes.
  arXiv  Detail & Related papers  (2023-08-27T06:32:20Z)
• Stroke-Based Autoencoders: Self-Supervised Learners for Efficient Zero-Shot Chinese Character Recognition [4.64065792373245]
  We develop a stroke-based autoencoder to model the sophisticated morphology of Chinese characters. Our SAE architecture outperforms other existing methods in zero-shot recognition.
  arXiv  Detail & Related papers  (2022-07-17T14:39:10Z)
• Few-Shot Font Generation by Learning Fine-Grained Local Styles [90.39288370855115]
  Few-shot font generation (FFG) aims to generate a new font with a few examples. We propose a new font generation approach by learning 1) the fine-grained local styles from references, and 2) the spatial correspondence between the content and reference glyphs.
  arXiv  Detail & Related papers  (2022-05-20T05:07:05Z)
• Font Completion and Manipulation by Cycling Between Multi-Modality Representations [113.26243126754704]
  We innovate to explore the generation of font glyphs as 2D graphic objects with the graph as an intermediate representation. We formulate a cross-modality cycled image-to-image structure with a graph between an image encoder and an image. Our model generates improved results than both image-to-image baseline and previous state-of-the-art methods for glyph completion.
  arXiv  Detail & Related papers  (2021-08-30T02:43:29Z)
• ZiGAN: Fine-grained Chinese Calligraphy Font Generation via a Few-shot Style Transfer Approach [7.318027179922774]
  ZiGAN is a powerful end-to-end Chinese calligraphy font generation framework. It does not require any manual operation or redundant preprocessing to generate fine-grained target-style characters. Our method has a state-of-the-art generalization ability in few-shot Chinese character style transfer.
  arXiv  Detail & Related papers  (2021-08-08T09:50:20Z)
• ChineseBERT: Chinese Pretraining Enhanced by Glyph and Pinyin Information [32.70080326854314]
  We propose ChineseBERT, which incorporates the glyph and pinyin information of Chinese characters into language model pretraining. The proposed ChineseBERT model yields significant performance boost over baseline models with fewer training steps.
  arXiv  Detail & Related papers  (2021-06-30T13:06:00Z)
• A Multi-Implicit Neural Representation for Fonts [79.6123184198301]
  font-specific discontinuities like edges and corners are difficult to represent using neural networks. We introduce textitmulti-implicits to represent fonts as a permutation-in set of learned implict functions, without losing features.
  arXiv  Detail & Related papers  (2021-06-12T21:40:11Z)
• Lattice-BERT: Leveraging Multi-Granularity Representations in Chinese Pre-trained Language Models [62.41139712595334]
  We propose a novel pre-training paradigm for Chinese -- Lattice-BERT. We construct a lattice graph from the characters and words in a sentence and feed all these text units into transformers. We show that our model can bring an average increase of 1.5% under the 12-layer setting.
  arXiv  Detail & Related papers  (2021-04-15T02:36:49Z)
• FGN: Fusion Glyph Network for Chinese Named Entity Recognition [5.653869962555973]
  We propose the FGN, Fusion Glyph Network for Chinese NER. FGN captures both glyph information and interactive information between glyphs from neighboring characters. Experiments are conducted on four NER datasets, showing that FGN with LSTM-CRF as tagger achieves new state-of-the-arts performance for Chinese NER.
  arXiv  Detail & Related papers  (2020-01-15T12:39:20Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.