In natural language processing (NLP), Part of Speech (POS) refers to the grammatical category or syntactic function that a word serves in a sentence. It's a way of categorizing words based on their roles within the structure of a sentence. POS tagging involves assigning a specific label, such as noun, verb, adjective, adverb, etc., to each word in a sentence.
Here are some common parts of speech:
| Tag | Arabic Tag | Description |
|---|---|---|
| Noun (N) | اسم | Represents a person, place, thing, or idea. Examples: dog, city, happiness. |
| Verb (V) | فعل | Describes an action or occurrence. Examples: run, eat, sleep. |
| Adjective (ADJ) | صفة | Modifies or describes a noun. Examples: happy, tall, red. |
| Adverb (ADV) | حال | Modifies or describes a verb, adjective, or other adverb. Examples: quickly, very, well. |
| Pronoun (PRON) | ضمير | Replaces a noun. Examples: he, she, it. |
| Preposition (PREP) | حرف جر | Indicates relationships between words, often in terms of time or place. Examples: in, on, under. |
| Conjunction (CONJ) | اقتران | Connects words, phrases, or clauses. Examples: and, but, or. |
| Interjection (INTJ) | تعجب | Expresses strong emotion. Examples: wow, oh, ouch. |
POS tagging is an essential task in NLP because understanding the grammatical structure of a sentence helps machines comprehend the meaning and context of the text. It's particularly useful in applications like text analysis, information retrieval, and language translation.
Tasneef.-.mp4
Demo link: https://tasneef.streamlit.app/
Hugging Face model card: https://huggingface.co/mohammedaly22/xlm-roberta-base-finetuned-ud-arabic
- Clone this repository
git clone https://github.com/MohammedAly22/Tasneef
- Inside the project folder, run the demo using the following command
streamlit run main.py
from transformers import pipeline
pos_tagger = pipeline("token-classification", "mohammedaly22/xlm-roberta-base-finetuned-ud-arabic")
text = "اشترى خالد سيارة، و أصبح يمتلك 3 سيارات."
predictions = pos_tagger(text)
words = [item["word"] for item in predictions]
predicted_entities = [item["entity"] for item in predictions]
print(f"words: {words}")
print(f"entites: {predicted_entities}")output:
words: ['▁اشتر', 'ى', '▁خالد', '▁سيارة', '،', '▁و', '▁أصبح', '▁يمتلك', '▁3', '▁سيارات', '.']
entites: ['VERB', 'VERB', 'X', 'NOUN', 'PUNCT', 'CCONJ', 'VERB', 'VERB', 'NUM', 'NOUN', 'PUNCT']
The Arabic-PADT UD treebank is based on the Prague Arabic Dependency Treebank (PADT), created at the Charles University in Prague.
The treebank consists of 7,664 sentences (282,384 tokens) and its domain is mainly newswire. The annotation is licensed under the terms of CC BY-NC-SA 3.0 and its original (non-UD) version can be downloaded from http://hdl.handle.net/11858/00-097C-0000-0001-4872-3.
The morphological and syntactic annotation of the Arabic UD treebank is created through the conversion of PADT data. The conversion procedure has been designed by Dan Zeman. The main coordinator of the original PADT project was Otakar Smrž.
Here is the first sample from the ar_padt-ud-test.conllu file:
# newdoc id = assabah.20041005.0017<br>
# newpar id = assabah.20041005.0017:p1<br>
# sent_id = assabah.20041005.0017:p1u1<br>
# text = <br>سوريا: تعديل وزاري واسع يشمل 8 حقائب
# orig_file_sentence ASB_ARB_20041005.0017#1<br>
1 سوريا سُورِيَا X X--------- Foreign=Yes 0 root 0:root SpaceAfter=No|Vform=سُورِيَا|Gloss=Syria|Root=sUr|Translit=sūriyā|LTranslit=sūriyā<br>
2 : : PUNCT G--------- _ 1 punct 1:punct Vform=:|Translit=:<br>
3 تعديل تَعدِيل NOUN N------S1I Case=Nom|Definite=Ind|Number=Sing 6 nsubj 6:nsubj<br> Vform=تَعدِيلٌ|Gloss=adjustment,change,modification,amendment|Root=_d_l|Translit=taʿdīlun|LTranslit=tadīl<br>
4 وزاري وِزَارِيّ ADJ A-----MS1I Case=Nom|Definite=Ind|Gender=Masc|Number=Sing 3 amod 3:amod<br> Vform=وِزَارِيٌّ|Gloss=ministry,ministerial|Root=w_z_r|Translit=wizārīyun|LTranslit=wizārīy<br>
5 واسع وَاسِع ADJ A-----MS1I Case=Nom|Definite=Ind|Gender=Masc|Number=Sing 3 amod 3:amod<br> Vform=وَاسِعٌ|Gloss=wide,extensive,broad|Root=w_s_|Translit=wāsiʿun|LTranslit=wāsi<br>
6 يشمل شَمِل VERB VIIA-3MS-- Aspect=Imp|Gender=Masc|Mood=Ind|Number=Sing|Person=3|VerbForm=Fin|Voice=Act 1 parataxis 1:parataxis<br> Vform=يَشمَلُ|Gloss=comprise,include,contain|Root=^s_m_l|Translit=yašmalu|LTranslit=šamil<br>
7 8 8 NUM Q--------- NumForm=Digit 6 obj 6:obj Vform=٨|Translit=8<br>
8 حقائب حَقِيبَة NOUN N------P2I Case=Gen|Definite=Ind|Number=Plur 7 nmod 7:nmod:gen<br> Vform=حَقَائِبَ|Gloss=briefcase,suitcase,portfolio,luggage|Root=.h_q_b|Translit=ḥaqāʾiba|LTranslit=ḥaqībat<br>
Here is the table of results after fine-tuning for 3 epochs:
| Epoch | Training Loss | Validation Loss | F1 | Accuracy |
|---|---|---|---|---|
| 1 | 0.188700 | 0.113953 | 0.958816 | 0.971454 |
| 2 | 0.090000 | 0.090725 | 0.966489 | 0.976796 |
| 3 | 0.055800 | 0.091719 | 0.969978 | 0.979393 |
confusion matrix:
