Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Choosing Transfer Languages for Cross-Lingual Learning(ACL 2019) #11

Open
akiFQC opened this issue Oct 21, 2019 · 0 comments
Open

Choosing Transfer Languages for Cross-Lingual Learning(ACL 2019) #11

akiFQC opened this issue Oct 21, 2019 · 0 comments
Labels
2019 published in 2019 ACL

Comments

@akiFQC
Copy link

akiFQC commented Oct 21, 2019

matome information

  • matome author: Fukuchi Akihiko
  • read date: 20191021

Choosing Transfer Languages for Cross-Lingual Learning

paper information

1. What did authors try to accomplish?

コーパスのサイズが小さいなどの理由によりマイナーな言語では、十分な学習精度をえいるためにタスクとは異なるコーパスサイズ大きな言語で事前に学習させ、その後にタ目的のタスク(言語)に転移学習させるという手法が知られている。しかし、ターミナルとなる言語に対して、どの言語から転移学習させるべきかはアドホックに決められていることが多い。そこで、異言語間での転移学習においてどの言語から転移させるべきかをヒューリスティックに決定する方法"LANGRANK"を提案した。

2. What were the key elements of the approach?

LANGRANKは、二言語間の特徴量からタスクの言語に対する複数additionalの言語との転移学習の相性をランキングで出力する。
LANGRANKの学習(評価)タスク:

  • machine translation (MT)
  • entity linking (EL)
  • part-of-speech (POS) tagging
  • dependency parsing (DEP)
    特徴量
  • Dataset Size
  • Type-Token Ratio (TTR): ユニークなtoken数の近さ
  • Word Overlap and Subword Overlap
  • Geographic distanc - Genetic distance
  • Inventory distanc : 音韻の普遍性
  • Syntactic distance - Phonological(音韻) distance
  • Featural distance : 上の5つを合わせたもの
    モデル: gradient boost dicision tree
    評価指標:Normalized Discounted Cumulative Gain (NDCG)

3. What can you use yourself?

試してみたい。
同じような方針で同一言語内でも事前学習やLMに使うコーパスを決定できそう。
LANGRANKで転移元言語が予測できない言語を調べると、比較言語学的に特殊な言語が見つかりそう!

4. What other references do you want to follow?

言語学に詳しくなさすぎて、特徴量の意味がよくわからない。
rankingの予測の知識がない。

その他メモ

github

official blog

  • なし

author's resource

  • なし

third person's explanation

  • なし

メモ

感想

cross-lingual transferに興味があったので、一番最初に手にとった論文を読んでみたら、かなり実用的内容で読んでいて充実感がありました。
転移元言語の選び方にタスク依存性があるのが気になりました。個人的には、言語とタスクをベクトル化して、両方の情報を用いてzero-shotで未知タスクに対して�転移元言語を選ぶというのも方向性としてあると思いました。

分野全体での立ち位置

@kirinsannnnnnnnnn kirinsannnnnnnnnn changed the title [paper title]([conference name][published year]) Choosing Transfer Languages for Cross-Lingual Learning(ACL 2019) Oct 21, 2019
@kirinsannnnnnnnnn kirinsannnnnnnnnn added 2019 published in 2019 ACL labels Oct 21, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
2019 published in 2019 ACL
Projects
None yet
Development

No branches or pull requests

2 participants