Add pythainlp.translate.word_translate

docker_requirements.txt

@@ -39,3 +39,4 @@ transformers==4.51.3
 ufal.chu-liu-edmonds==1.0.3
 wtpsplit==1.3.0
 wunsen==0.0.3
+word2word==1.0.0

docs/api/translate.rst

@@ -12,6 +12,9 @@ Modules
 
    The `Translate` class is the central component of the module, offering a unified interface for various translation tasks. It acts as a coordinator, directing translation requests to specific language pairs and models.
 
+.. autoclass:: word_translate
+   :members:
+
 .. autofunction::  pythainlp.translate.en_th.download_model_all
     :noindex:
 

pythainlp/translate/__init__.py

@@ -6,9 +6,9 @@
 Language translation.
 """
 
-__all__ = ["Translate", "ThZhTranslator", "ZhThTranslator"]
+__all__ = ["Translate", "ThZhTranslator", "ZhThTranslator", "word_translate"]
 
-from pythainlp.translate.core import Translate
+from pythainlp.translate.core import Translate, word_translate
 from pythainlp.translate.zh_th import (
     ThZhTranslator,
     ZhThTranslator,

pythainlp/translate/core.py

@@ -2,6 +2,8 @@
 # SPDX-FileCopyrightText: 2016-2025 PyThaiNLP Project
 # SPDX-FileType: SOURCE
 # SPDX-License-Identifier: Apache-2.0
+from typing import List, Union
+
 
 class Translate:
     """
@@ -83,7 +85,7 @@ def load_model(self):
         else:
             raise ValueError("Not support language!")
 
-    def translate(self, text) -> str:
+    def translate(self, text: str) -> str:
         """
         Translate text
 
@@ -94,3 +96,43 @@ def translate(self, text) -> str:
         if self.engine == "small100":
             return self.model.translate(text, tgt_lang=self.target_lang)
         return self.model.translate(text)
+
+
+def word_translate(
+        word: str,
+        src: str,
+        target: str,
+        engine: str="word2word"
+    ) -> Union[List[str], None]:
+    """
+    Translate word from source language to target language.
+
+    :param str word: text
+    :param str src: src language
+    :param str target: target language
+    :param str engine: Word translate engine (the default engine is word2word)
+    :return: return list word translate or None
+    :rtype: Union[List[str], None]
+
+    :Example:
+
+    Translate word from Thai to English::
+
+        from pythainlp.translate import word_translate
+        print(word_translate("แมว","th","en"))
+        # output: ['cat', 'cats', 'kitty', 'kitten', 'Cat']
+
+    Translate word from English to Thai::
+
+        from pythainlp.translate import word_translate
+        print(word_translate("cat","en","th"))
+        # output: ['แมว', 'แมวป่า', 'ข่วน', 'เลี้ยง', 'อาหาร']
+
+    """
+    if engine=="word2word":
+        from .word2word_translate import translate
+        return translate(word=word, src=src, target=target)
+    else:
+        raise NotImplementedError(
+            f"pythainlp.translate.word_translate isn't support {engine}."
+        )

pythainlp/translate/th_fr.py

@@ -13,7 +13,6 @@
 
 - Huggingface https://huggingface.co/Helsinki-NLP/opus-mt-th-fr
 """
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 
 
 class ThFrTranslator:
@@ -36,6 +35,7 @@ def __init__(
         use_gpu: bool = False,
         pretrained: str = "Helsinki-NLP/opus-mt-th-fr",
     ) -> None:
+        from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
         self.tokenizer_thzh = AutoTokenizer.from_pretrained(pretrained)
         self.model_thzh = AutoModelForSeq2SeqLM.from_pretrained(pretrained)
         if use_gpu:

pythainlp/translate/word2word_translate.py

@@ -0,0 +1,87 @@
+# -*- coding: utf-8 -*-
+# SPDX-FileCopyrightText: 2016-2025 PyThaiNLP Project
+# SPDX-FileType: SOURCE
+# SPDX-License-Identifier: Apache-2.0
+from typing import List, Union
+from word2word import Word2word
+
+support_list = set(['zh_tw',
+ 'el',
+ 'te',
+ 'hu',
+ 'eu',
+ 'ko',
+ 'ru',
+ 'lv',
+ 'bg',
+ 'sk',
+ 'vi',
+ 'gl',
+ 'et',
+ 'ta',
+ 'fa',
+ 'it',
+ 'ms',
+ 'id',
+ 'pt',
+ 'fr',
+ 'sr',
+ 'mk',
+ 'sv',
+ 'si',
+ 'en',
+ 'ka',
+ 'uk',
+ 'sl',
+ 'hi',
+ 'ca',
+ 'lt',
+ 'es',
+ 'no',
+ 'de',
+ 'he',
+ 'cs',
+ 'ze_zh',
+ 'fi',
+ 'pl',
+ 'tl',
+ 'is',
+ 'ze_en',
+ 'kk',
+ 'bn',
+ 'tr',
+ 'ur',
+ 'pt_br',
+ 'ar',
+ 'ro',
+ 'bs',
+ 'ml',
+ 'zh_cn',
+ 'da',
+ 'hr',
+ 'sq',
+ 'af',
+ 'eo',
+ 'nl',
+ 'ja',
+ 'th'])
+
+
+def translate(word: str, src: str, target: str) -> Union[List[str], None]:
+    """
+    Word translate
+
+    :param str word: text
+    :param str src: src language
+    :param str target: target language
+    :return: return list word translate or None
+    :rtype: Union[List[str], None]
+    """
+    if src not in support_list or target not in support_list:
+        raise NotImplementedError(
+            f"word2word doesn't support {src}-{target}."
+        )
+    elif src==target:
+        return [word]
+    _engine = Word2word(src, target)
+    return _engine(word)

pythainlp/translate/zh_th.py

@@ -10,7 +10,6 @@
 - GitHub: https://github.com/LalitaDeelert/lalita-mt-zhth
 - Facebook post https://web.facebook.com/aibuildersx/posts/166736255494822
 """
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 
 
 class ThZhTranslator:
@@ -30,6 +29,7 @@ def __init__(
         use_gpu: bool = False,
         pretrained: str = "Lalita/marianmt-th-zh_cn",
     ) -> None:
+        from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
         self.tokenizer_thzh = AutoTokenizer.from_pretrained(pretrained)
         self.model_thzh = AutoModelForSeq2SeqLM.from_pretrained(pretrained)
         if use_gpu:

setup.py

@@ -85,6 +85,7 @@
         "sentencepiece>=0.1.91",
         "torch>=1.0.0",
         "transformers>=4.6.0",
+        "word2word>=1.0.0"
     ],
     "transformers_ud": [
         "transformers>=4.22.1",
@@ -145,6 +146,7 @@
         "ufal.chu-liu-edmonds>=1.0.2",
         "wtpsplit>=1.0.1",
         "wunsen>=0.0.3",
+        "word2word>=1.0.0",
     ],
 }
 

tests/extra/testx_translate.py

@@ -5,7 +5,7 @@
 
 import unittest
 
-from pythainlp.translate import Translate
+from pythainlp.translate import Translate, word_translate
 from pythainlp.translate.en_th import (
     EnThTranslator,
     ThEnTranslator,
@@ -73,3 +73,23 @@ def test_translate(self):
         # )
         with self.assertRaises(ValueError):
             self.th_cat_translator = Translate('th', 'cat', engine="fkfj")
+
+    def test_word_translate(self):
+        self.assertIsNone(word_translate("cat", src="en", target="th"))
+        self.assertIsNone(word_translate("แมว", src="en", target="th"))
+        self.assertIsNone(
+            word_translate("cat", src="en", target="th", engine="word2word")
+        )
+        self.assertIsNone(
+            word_translate("แมว", src="en", target="th", engine="word2word")
+        )
+        self.assertEqual(
+            word_translate("แมว", src="th", target="th", engine="word2word"),
+            ["แมว"]
+        )
+
+        with self.assertRaises(NotImplementedError):
+            word_translate("cat", src="en", target="th", engine="cat")
+
+        with self.assertRaises(NotImplementedError):
+            word_translate("แมว", src="th", target="xxx", engine="word2word")