Awesome Open Source
Awesome Open Source

OpenHowNet Logo


This project contains core data of HowNet and OpenHowNet API developed by THUNLP, which provides a convenient way to search information in HowNet, display sememe trees, calculate word similarity via sememes, etc. You can also visit our website to enjoy searching and exhibiting sememes of words online.

If you use any data or API provided by OpenHowNet in your research, please cite the following paper:

  title={OpenHowNet: An Open Sememe-based Lexical Knowledge Base},
  author={Qi, Fanchao and Yang, Chenghao and Liu, Zhiyuan and Dong, Qiang and Sun, Maosong and Dong, Zhendong},
  journal={arXiv preprint arXiv:1901.09957},

HowNet Core Data

HowNet core data file(HowNet.txt)consists of concepts represented by 223,767 Chinese & English words and phrases. Each concept in HowNet is annotated with sememe-based definition, POS tag, sentiment oriatation, examples, etc. Here is an example of how concepts are annotated in HowNet:

NO.=042012 #Concept ID
W_C=贷 #Chinese word
G_C=verb [9MustObj] [dai4] #POS tag of the Chinese word
S_C=PlusFeeling|正面情感 #Sentiment orientation
E_C=定斩不~,严惩不~  #Examples of the Chinese word
W_E=forgive #English word 
G_E=verb [7 forgiveverb-0vt,sobj,ofnpa22    ]  #POS tag of the English word
S_E=PlusFeeling|正面情感 #Sentiment orientation
E_E=    #Examples of the English word
DEF={forgive|原谅} # Sememe-based definition

OpenHowNet API


  • Python==3.6
  • anytree==2.4.3
  • tqdm==4.31.1
  • requests==2.22.0


  • Installation via pip (recommended)
pip install OpenHowNet
  • Installation via Github
git clone
cd OpenHowNet/OpenHowNet


interfaces description params
get(self, word, language=None) Search all information annotated with a word. word is the target word. lang is en(English) or zh(Chinese), searching in both languages by default.
get_sememes_by_word(self, word, structured=False, lang='zh', merge=False, expanded_layer=-1) Search sememes of the target word. You can choose whether multiple senses in the result are merged, whether the result itself is structured, and the expand layers of the tree. word is the target word. lang is en(English) or zh(Chinese). structured denotes whether the result is structured. merge denotes whether the result is merged. expanded_layer is number of expanded layers and -1 means expand all layers.
initialize_sememe_similarity_calculation(self) Initialize the implementation of advanced feature to calculate sememe-based word similarity. It may take some time to read necessary files.
calculate_word_similarity(self, word0, word1) Calculate similarity of two words. You need to run last initialization command before calling this function. word0 and word1 represents the words you are querying.
get_nearest_words_via_sememes(self, word, K=10) Get the nearest K words to the target word with similarity calculated via sememes. word is the target word, K is "top K" in k nearest neighbors.
get_sememe_relation(self, sememe0, sememe1) Get the relationship between two sememes. sememe0 and sememe1 represent the sememes you are querying.
get_sememe_via_relation(self, sememe, relation, lang='zh') Get all sememes that have specified relation with the input sememe. sememe is the target sememe, relation is the target relation, lang is en(English) or zh(Chinese).



import OpenHowNet
hownet_dict = OpenHowNet.HowNetDict()

An error will occur if you haven't downloaded the HowNet data. In this case you need to run first.

Get Word Annotations in HowNet

By default, the api will search the target word in both English and Chinese annotations in HowNet, which will cause significant search overhead. Note that if the target word does not exist in HowNet annotation, this api will simply return an empty list.

>>> result_list = hownet_dict.get("苹果")
>>> print(len(result_list))
>>> print(result_list[0])
{'Def': '{computer|电脑:modifier={PatternValue|样式值:CoEvent={able|能:scope={bring|携带:patient={$}}}}{SpeBrand|特定牌子}}', 'en_grammar': 'noun', 'zh_grammar': 'noun', 'No': '127151', 'syn': [{'id': '004024', 'text': 'IBM'}, {'id': '041684', 'text': '戴尔'}, {'id': '049006', 'text': '东芝'}, {'id': '106795', 'text': '联想'}, {'id': '156029', 'text': '索尼'}, {'id': '004203', 'text': 'iPad'}, {'id': '019457', 'text': '笔记本'}, {'id': '019458', 'text': '笔记本电脑'}, {'id': '019459', 'text': '笔记本电脑'}, {'id': '019460', 'text': '笔记本电脑'}, {'id': '019461', 'text': '笔记本电脑'}, {'id': '019463', 'text': '笔记簿电脑'}, {'id': '019464', 'text': '笔记簿电脑'}, {'id': '020567', 'text': '便携式电脑'}, {'id': '020568', 'text': '便携式计算机'}, {'id': '020569', 'text': '便携式计算机'}, {'id': '127224', 'text': '平板电脑'}, {'id': '127225', 'text': '平板电脑'}, {'id': '172264', 'text': '膝上型电脑'}, {'id': '172265', 'text': '膝上型电脑'}], 'zh_word': '苹果', 'en_word': 'apple'}

>>> hownet_dict.get("test_for_non_exist_word")

You can visualize the retrieved HowNet structured sememe annotations ("sememe tree") of the target word as follows (K=2 means only displaying sememe trees of 2 concepts represented by the input word)

>>> hownet_dict.visualize_sememe_trees("苹果", K=2)
Find 6 result(s)
Display #0 sememe tree
└── [None]computer|电脑
    ├── [modifier]PatternValue|样式值
    │   └── [CoEvent]able|能
    │       └── [scope]bring|携带
    │           └── [patient]$
    └── [patient]SpeBrand|特定牌子
Display #1 sememe tree
└── [None]fruit|水果

To boost the efficiency of the search process, you can specify the language of the target word as the following.

>>> result_list = hownet_dict.get("苹果", language="zh")
>>> print("Number of monolingual results: ",len(result_list))
Number of monolingual results: 6
>>> print("Example of monolingual results: ",result_list[0])
Example of monolingual results: {'Def': '{computer|电脑:modifier={PatternValue|样式值:CoEvent={able|能:scope={bring|携带:patient={$}}}}{SpeBrand|特定牌子}}', 'en_grammar': 'noun', 'zh_grammar': 'noun', 'No': '127151', 'syn': [{'id': '004024', 'text': 'IBM'}, {'id': '041684', 'text': '戴尔'}, {'id': '049006', 'text': '东芝'}, {'id': '106795', 'text': '联想'}, {'id': '156029', 'text': '索尼'}, {'id': '004203', 'text': 'iPad'}, {'id': '019457', 'text': '笔记本'}, {'id': '019458', 'text': '笔记本电脑'}, {'id': '019459', 'text': '笔记本电脑'}, {'id': '019460', 'text': '笔记本电脑'}, {'id': '019461', 'text': '笔记本电脑'}, {'id': '019463', 'text': '笔记簿电脑'}, {'id': '019464', 'text': '笔记簿电脑'}, {'id': '020567', 'text': '便携式电脑'}, {'id': '020568', 'text': '便携式计算机'}, {'id': '020569', 'text': '便携式计算机'}, {'id': '127224', 'text': '平板电脑'}, {'id': '127225', 'text': '平板电脑'}, {'id': '172264', 'text': '膝上型电脑'}, {'id': '172265', 'text': '膝上型电脑'}], 'zh_word': '苹果', 'en_word': 'apple'}

>>> print("Number of all the results: ",len(hownet_dict.get("X")))
Number of all the results: 5
>>> print("Number of Chinese results: ",len(hownet_dict.get("X",language="zh")))
Number of Chinese results: 3
>>> print("Number of English results:",len(hownet_dict.get("X",language="en")))
Number of English results: 2

>>> hownet_dict.get("苹果", language="en")

Get All Words Annotated in HowNet

>>> zh_word_list = hownet_dict.get_zh_words()
>>> print(zh_word_list[:30])
['', '"', '#', '#号标签', '$', '%', "'", '(', ')', '*', '+', '-', '--', '...', '...出什么问题', '...底', '...底下', '...发生故障', '...发生了什么', '...何如', '...家里有几口人', '...检测呈阳性', '...检测呈阴性', '...来', '...内', '...为止', '...也同样使然', '...以来', '...以内', '...以上']

>>> en_word_list = hownet_dict.get_en_words()
>>> print(en_word_list[:30])
['A', 'An', 'Frenchmen', 'Frenchwomen', 'Ottomans', 'a', 'aardwolves', 'abaci', 'abandoned', 'abbreviated', 'abode', 'aboideaux', 'aboiteaux', 'abscissae', 'absorbed', 'acanthi', 'acari', 'accepted', 'acciaccature', 'acclaimed', 'accommodating', 'accompanied', 'accounting', 'accused', 'acetabula', 'acetified', 'aching', 'acicula', 'acini', 'acquired']

Get Flattened Sememe Trees for the Input Word

Notice: the parameters lang, merge and expanded_layer only work when structured = False. The main consideration is that there are multiple ways to interpret these params when deal with structured data. We leave the freedom to our end user. In next section, you will be able to see how to utilize the structured data. Detailed descriptions of the params are in our documentation.

Get the full merged sememe list from multi-sense words

>>> hownet_dict.get_sememes_by_word("苹果",structured=False,lang="zh",merge=True)
{'电脑', '交流', '用具', '水果', '特定牌子', '样式值', '能', '树', '生殖', '携带'}

>>> hownet_dict.get_sememes_by_word("apple",structured=False,lang="en",merge=True)
{'communicate', 'able', 'reproduce', 'SpeBrand', 'computer', 'bring', 'tool', 'PatternValue', 'tree', '$', 'fruit'}

Even if the language is not corresponding to the target word, the api still works. It will keep all the returned word entries to be in the same language you specified.

>>> hownet_dict.get_sememes_by_word("苹果",structured=False,lang="en",merge=True)
{'apple': {'communicate', 'able', 'reproduce', 'SpeBrand', 'computer', 'bring', 'tool', 'PatternValue', 'tree', '$', 'fruit'}, 'malus pumila': {'reproduce', 'fruit', 'tree'}, 'orchard apple tree': {'reproduce', 'fruit', 'tree'}}

You could specify the number of the expanded layers:

>>> hownet_dict.get_sememes_by_word("苹果",structured=False,merge=True,expanded_layer=2)
{'电脑', '树', '用具', '水果'}

You could get all flattened sememe trees for all words as well as specify the number of the expanded layers:

>>> hownet_dict.get_sememes_by_word("I WANT ALL!",structured=False,merge=True)
# the result is too large, just try it yourself.

If you would like to see the sememe lists for different senses of particular word in HowNet, just need to set the param merged to False.

>>> hownet_dict.get_sememes_by_word("苹果",structured=False,lang="zh",merge=False)
[{'word': '苹果', 'sememes': {'特定牌子', '样式值', '电脑', '能', '携带'}},
{'word': '苹果', 'sememes': {'水果'}},
{'word': '苹果', 'sememes': {'特定牌子', '样式值', '能', '交流', '用具', '携带'}},
{'word': '苹果', 'sememes': {'树', '生殖', '水果'}},
{'word': '苹果', 'sememes': {'树', '生殖', '水果'}},
{'word': '苹果', 'sememes': {'树', '生殖', '水果'}}]

>>> hownet_dict.get_sememes_by_word("apple",structured=False,lang="en",merge=False)
[{'word': 'apple', 'sememes': {'able', 'computer', 'bring', 'SpeBrand', 'PatternValue', '$'}},
{'word': 'apple', 'sememes': {'fruit'}},
{'word': 'apple', 'sememes': {'communicate', 'able', 'bring', 'tool', 'SpeBrand', 'PatternValue', '$'}},
{'word': 'apple', 'sememes': {'reproduce', 'fruit', 'tree'}},
{'word': 'apple', 'sememes': {'communicate', 'able', 'bring', 'tool', 'SpeBrand', 'PatternValue', '$'}},
{'word': 'apple', 'sememes': {'reproduce', 'fruit', 'tree'}},
{'word': 'apple', 'sememes': {'fruit'}},
{'word': 'apple', 'sememes': {'fruit'}}]

Get Structured Sememe Trees for Certain Words in HowNet

>>> hownet_dict.get_sememes_by_word("苹果",structured=True)[0]["tree"]
{'role': 'sense', 'name': '苹果','children': [
    {'role': 'None', 'name': 'computer|电脑', 'children': [
        {'role': 'modifier', 'name': 'PatternValue|样式值', 'children': [
            {'role': 'CoEvent', 'name': 'able|能', 'children': [
                {'role': 'scope', 'name': 'bring|携带', 'children': [
                    {'role': 'patient', 'name': '$'}
        {'role': 'patient', 'name': 'SpeBrand|特定牌子'}

Two ways to see the corresponding annotation data

>>> hownet_dict.get_sememes_by_word("苹果",structured=True)[0]["tree"] # or
>>> hownet_dict.get_sememes_by_word("苹果",structured=True)[0]["word"]
>>> # two results are the same, only displaying one
{'Def': '{computer|电脑:modifier={PatternValue|样式值:CoEvent={able|能:scope={bring|携带:patient={$}}}}{SpeBrand|特定牌子}}',
'en_grammar': 'noun',
'zh_grammar': 'noun',
'No': '127151',
'syn': [
    {'id': '004024', 'text': 'IBM'},
    {'id': '041684', 'text': '戴尔'},
    {'id': '049006', 'text': '东芝'},
    {'id': '106795', 'text': '联想'},
    {'id': '156029', 'text': '索尼'},
    {'id': '004203', 'text': 'iPad'},
    {'id': '019457', 'text': '笔记本'},
    {'id': '019458', 'text': '笔记本电脑'},
    {'id': '019459', 'text': '笔记本电脑'},
    {'id': '019460', 'text': '笔记本电脑'},
    {'id': '019461', 'text': '笔记本电脑'},
    {'id': '019463', 'text': '笔记簿电脑'},
    {'id': '019464', 'text': '笔记簿电脑'},
    {'id': '020567', 'text': '便携式电脑'},
    {'id': '020568', 'text': '便携式计算机'},
    {'id': '020569', 'text': '便携式计算机'},
    {'id': '127224', 'text': '平板电脑'},
    {'id': '127225', 'text': '平板电脑'},
    {'id': '172264', 'text': '膝上型电脑'},
    {'id': '172265', 'text': '膝上型电脑'}
'zh_word': '苹果',
'en_word': 'apple'}

Get the Synonyms of the Input Word

The similarity metrics are based on sememes.

>>> hownet_dict["苹果"][0]["syn"]
[{'id': '004024', 'text': 'IBM'},
 {'id': '041684', 'text': '戴尔'},
 {'id': '049006', 'text': '东芝'},
 {'id': '106795', 'text': '联想'},
 {'id': '156029', 'text': '索尼'},
 {'id': '004203', 'text': 'iPad'},
 {'id': '019457', 'text': '笔记本'},
 {'id': '019458', 'text': '笔记本电脑'},
 {'id': '019459', 'text': '笔记本电脑'},
 {'id': '019460', 'text': '笔记本电脑'},
 {'id': '019461', 'text': '笔记本电脑'},
 {'id': '019463', 'text': '笔记簿电脑'},
 {'id': '019464', 'text': '笔记簿电脑'},
 {'id': '020567', 'text': '便携式电脑'},
 {'id': '020568', 'text': '便携式计算机'},
 {'id': '020569', 'text': '便携式计算机'},
 {'id': '127224', 'text': '平板电脑'},
 {'id': '127225', 'text': '平板电脑'},
 {'id': '172264', 'text': '膝上型电脑'},
 {'id': '172265', 'text': '膝上型电脑'}]

Query a Word by its ID

>>> hownet_dict["004024"]
['Def', 'en_grammar', 'zh_grammar', 'No', 'syn', 'zh_word', 'en_word']

Get all sememes

>>> len(hownet_dict.get_all_sememes())

Get Relationship Between Two Sememes

The sememes you input can be in any language.

>>> hownet_dict.get_sememe_relation("音量值", "尖声")

>>> hownet_dict.get_sememe_relation("尖声", "SoundVolumeValue")

>>> hownet_dict.get_sememe_relation("shrill", "SoundVolumeValue")

>>> hownet_dict.get_sememe_relation("音量值", "shrill")

The output could be hypernym, hyponym, antonym or converse.

Get sememes having a certain relation with the input sememe

The sememe you input can be in any language, but the relation must be in lowercase English. You can specify the language of result, by default it will be Chinese.

>>> hownet_dict.get_sememe_via_relation("音量值", "hyponym")
['高声', '低声', '尖声', '沙哑', '无声', '有声']

>>> hownet_dict.get_sememe_via_relation("音量值", "hyponym", lang="en")
['loud', 'LowVoice', 'shrill', 'hoarse', 'silent', 'talking']

>>> hownet_dict.get_sememe_via_relation("SoundVolumeValue", "hyponym", lang="en")
['loud', 'LowVoice', 'shrill', 'hoarse', 'silent', 'talking']

Advanced Feature: Word Similarity Calculation via Sememes

Our implementation is based on the paper:

Jiangming Liu, Jinan Xu, Yujie Zhang. An Approach of Hybrid Hierarchical Structure for Word Similarity Computing by HowNet. In Proceedings of IJCNLP

Extra Initialization

Because there are some files required to be loaded for similarity calculation, the initialization overhead will be larger than before. To begin with, you can initialize the hownet_dict object as the following code:

>>> hownet_dict_advanced = OpenHowNet.HowNetDict(use_sim=True)

You can also postpone the initialization work of similarity calculation until use. The following code serves as an example and the return value will indicate whether the extra initialization process succeed.

>>> hownet_dict.initialize_sememe_similarity_calculation()

Get Top-K Nearest Words for the Input Word

If the given word does not exist in HowNet annotations, this function will return an empty list.

>>> query_result = hownet_dict_advanced.get_nearest_words_via_sememes("苹果",20)
>>> example = query_result[0]
>>> print("word_name:",example["word"])
>>> print("id:",example["id"])
>>> print("synset and corresonding word&id&score:")
>>> print(example["synset"])
word_name: 苹果
id: 127151
synset and corresonding word&id&score:
[{'id': 4024, 'word': 'IBM', 'score': 1.0},
 {'id': 41684, 'word': '戴尔', 'score': 1.0},
 {'id': 49006, 'word': '东芝', 'score': 1.0},
 {'id': 106795, 'word': '联想', 'score': 1.0},
 {'id': 156029, 'word': '索尼', 'score': 1.0},
 {'id': 4203, 'word': 'iPad', 'score': 0.865},
 {'id': 19457, 'word': '笔记本', 'score': 0.865},
 {'id': 19458, 'word': '笔记本电脑', 'score': 0.865},
 {'id': 19459, 'word': '笔记本电脑', 'score': 0.865},
 {'id': 19460, 'word': '笔记本电脑', 'score': 0.865},
 {'id': 19461, 'word': '笔记本电脑', 'score': 0.865},
 {'id': 19463, 'word': '笔记簿电脑', 'score': 0.865},
 {'id': 19464, 'word': '笔记簿电脑', 'score': 0.865},
 {'id': 20567, 'word': '便携式电脑', 'score': 0.865},
 {'id': 20568, 'word': '便携式计算机', 'score': 0.865},
 {'id': 20569, 'word': '便携式计算机', 'score': 0.865},
 {'id': 127224, 'word': '平板电脑', 'score': 0.865},
 {'id': 127225, 'word': '平板电脑', 'score': 0.865},
 {'id': 172264, 'word': '膝上型电脑', 'score': 0.865},
 {'id': 172265, 'word': '膝上型电脑', 'score': 0.865}]

Calculate the Similarity for Given Two Words

If any of the given words does not exist in HowNet annotations, this function will return 0.

>>> hownet_dict_advanced.calculate_word_similarity("苹果", "梨")

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
Jupyter Notebook (231,353
Nlp (7,908
Semantic (472
Knowledge Base (361
Related Projects