API Reference

This section provides complete API documentation for all Underthesea functions.

Core Functions

Function	Description	Install
sent_tokenize	Sentence segmentation	Core
text_normalize	Text normalization	Core
word_tokenize	Word segmentation	Core
pos_tag	Part-of-speech tagging	Core
chunk	Phrase chunking	Core
ner	Named entity recognition	Core
classify	Text classification	Core
sentiment	Sentiment analysis	Core
convert_address	Address conversion (63→34 provinces)	Core

Deep Learning Functions

Function	Description	Install
dependency_parse	Dependency parsing	[deep]
translate	Vietnamese-English translation	[deep]

Additional Functions

Function	Description	Install
lang_detect	Language detection	[langdetect]
tts	Text-to-speech	[voice]
agent	Conversational AI agent	[agent]

Quick Import

All main functions can be imported directly from underthesea:

from underthesea import (
    sent_tokenize,
    text_normalize,
    word_tokenize,
    pos_tag,
    chunk,
    ner,
    classify,
    sentiment,
    dependency_parse,  # requires [deep]
    translate,         # requires [deep]
    lang_detect,       # requires [langdetect]
    agent,             # requires [agent]
    convert_address,
)

Common Parameters

Many functions share common parameters:

format

Controls output format:

• None (default): Returns a list
• "text": Returns a string with underscores joining multi-word tokens

word_tokenize("Việt Nam", format=None)   # ['Việt Nam']
word_tokenize("Việt Nam", format="text") # 'Việt_Nam'

model

Specifies which model to use:

# Use default model
ner("text")

# Use specific model
ner("text", deep=True)  # Use deep learning model
classify("text", model='prompt')  # Use OpenAI model

domain

Specifies the domain for domain-specific models:

classify("text", domain='bank')
sentiment("text", domain='bank')