Code Switch

CodeSwitch is a NLP tool, can use for language identification, pos tagging, name entity recognition, sentiment analysis of code mixed data.

Supported Code-Mixed Language

We used LinCE dataset for training multilingual BERT model using huggingface transformers. LinCE has four language mixed data. We took three of it spanish-english, hindi-english and nepali-english. Hope we will train and add other language and task too.

• Spanish-English(spa-eng)
• Hindi-English(hin-eng)
• Nepali-English(nep-eng)

Language Code

• spa-eng for spanish-english
• hin-eng for hindi-english
• nep-eng for nepali-english

Installation

pip install codeswitch

Dependency

• pytorch >=1.6.0

Training Details

• All three(lid, ner, pos) sequence tagging model was trainend with huggingface token classification
• Sentiment Analysis Model trained with huggingface text classification
• You can find every model and evaluation results here

Features & Supported Language

• Language Identification
  • spanish-english
  • hindi-english
  • nepali-english
• POS
  • spanish-english
  • hindi-english
• NER
  • spanish-english
  • hindi-english
• Sentiment Analysis
  • spanish-english

Language Identification

from codeswitch.codeswitch import LanguageIdentification
lid = LanguageIdentification('spa-eng') 
# for hindi-english use 'hin-eng', 
# for nepali-english use 'nep-eng'
text = "" # your code-mixed sentence 
result = lid.identify(text)
print(result)

POS Tagging

from codeswitch.codeswitch import POS
pos = POS('spa-eng')
# for hindi-english use 'hin-eng'
text = "" # your mixed sentence 
result = pos.tag(text)
print(result)

NER Tagging

from codeswitch.codeswitch import NER
ner = NER('spa-eng')
# for hindi-english use 'hin-eng'
text = "" # your mixed sentence 
result = ner.tag(text)
print(result)

Sentiment Analysis

from codeswitch.codeswitch import SentimentAnalysis
sa = SentimentAnalysis('spa-eng')
sentence = "El perro le ladraba a La Gatita .. .. lol #teamlagatita en las playas de Key Biscayne este Memorial day"
result = sa.analyze(sentence)
print(result)
# [{'label': 'LABEL_1', 'score': 0.9587041735649109}]

Acknowledgement

• LinCE
• BERT
• huggingface