search

UKPLab / elmo-bilstm-cnn-crf

BiLSTM-CNN-CRF architecture for sequence tagging using ELMo representations.
Apache License 2.0
389 stars 81 forks source link

Where is the cnn layer? #30

Open qiangailin opened 4 years ago

qiangailin commented 4 years ago

Hello, the model architecture is bilstm-cnn-crf, I want to know where the cnn layer is？

nreimers commented 4 years ago

Can you explain what you mean?

The architecture is described in this paper: https://arxiv.org/abs/1603.01354

The CNN layer ist used to derive from the characters of a word a word representation, that is then feed into a BiLSTM layer, followed by a CRF layer.

With ELMo, this CNN layer is not really needed, as ELMo is already using a CNN on character-level to derive word representations.

Best regards Nils Reimers

qiangailin commented 4 years ago

The explanation is very clear,thanks

------------------ 原始邮件 ------------------ 发件人: "Nils Reimers"<notifications@github.com>; 发送时间: 2019年10月23日(星期三) 上午6:30 收件人: "UKPLab/elmo-bilstm-cnn-crf"<elmo-bilstm-cnn-crf@noreply.github.com>; 抄送: "凌霄大将"<971641064@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [UKPLab/elmo-bilstm-cnn-crf] Where is the cnn layer? (#30)

Can you explain what you mean?

The architecture is described in this paper: https://arxiv.org/abs/1603.01354

The CNN layer ist used to derive from the characters of a word a word representation, that is then feed into a BiLSTM layer, followed by a CRF layer.

With ELMo, this CNN layer is not really needed, as ELMo is already using a CNN on character-level to derive word representations.

Best regards Nils Reimers

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.