In this paper, we propose new methods to learn Chinese word representations.Chinese characters are composed of graphical components, which carry richsemantics. It is common for a Chinese learner to comprehend the meaning of aword from these graphical components. As a result, we propose models thatenhance word representations by character glyphs. The character glyph featuresare directly learned from the bitmaps of characters by convolutionalauto-encoder(convAE), and the glyph features improve Chinese wordrepresentations which are already enhanced by character embeddings. Anothercontribution in this paper is that we created several evaluation datasets intraditional Chinese and made them public.
展开▼