--- tags: - flair - token-classification - sequence-tagger-model language: fr widget: - text: "George Washington est allé à Washington" --- **People Involved** * LABRAK Yanis (1) * DUFOUR Richard (2) **Affiliations** 1. LIA, Avignon University, Avignon, France. 2. LS2N, Nantes University, Nantes, France. # POET: A French Extended Part-of-Speech Tagger - Corpus: [UD_FRENCH_TREEBANKS](https://universaldependencies.org/treebanks/fr_gsd/index.html) - Model: [Flair](https://www.aclweb.org/anthology/C18-1139.pdf) - Embeddings: [FastText](https://fasttext.cc/) - Additionnel: [LSTM-CRF](https://arxiv.org/abs/1011.4088) - Nombre d'Epochs: 115 ## Demo: How to use in Flair Requires [Flair](https://pypi.org/project/flair/): ```pip install flair``` ```python from flair.data import Sentence from flair.models import SequenceTagger # Load the model model = SequenceTagger.load("qanastek/pos-french") sentence = Sentence("George Washington est allé à Washington") # predict tags model.predict(sentence) # print predicted pos tags print(sentence.to_tagged_string()) ``` Output: ![Preview Output](preview.PNG) ## Training data `UD_FRENCH_GSD_Plus` is a part-of-speech tagging corpora based on [UD_French-GSD](https://universaldependencies.org/treebanks/fr_gsd/index.html) which was originally created in 2015 and is based on the [universal dependency treebank v2.0](https://github.com/ryanmcd/uni-dep-tb). Originally, the corpora consists of 400,399 words (16,341 sentences) and had 17 different classes. Now, after applying our tags augmentation we obtain 60 different classes which add semantic information such as the gender, number, mood, person, tense or verb form given in the different CoNLL-03 fields from the original corpora. We based our tags on the level of details given by the [LIA_TAGG](http://pageperso.lif.univ-mrs.fr/frederic.bechet/download.html) statistical POS tagger written by [Frédéric Béchet](http://pageperso.lif.univ-mrs.fr/frederic.bechet/index-english.html) in 2001. ## Original Tags ```plain PRON VERB SCONJ ADP CCONJ DET NOUN ADJ AUX ADV PUNCT PROPN NUM SYM PART X INTJ ``` ## New Tags | Abbreviation | Description | Examples | |:--------:|:--------:|:--------:| | PREP | Preposition | de | | AUX | Auxiliary Verb | est | | ADV | Adverb | toujours | | COSUB | Subordinating conjunction | que | | COCO | Coordinating Conjunction | et | | PART | Demonstrative particle | -t | | PRON | Pronoun | qui ce quoi | | PDEMMS | Singular Masculine Demonstrative Pronoun | ce | | PDEMMP | Plurial Masculine Demonstrative Pronoun | ceux | | PDEMFS | Singular Feminine Demonstrative Pronoun | cette | | PDEMFP | Plurial Feminine Demonstrative Pronoun | celles | | PINDMS | Singular Masculine Indefinite Pronoun | tout | | PINDMP | Plurial Masculine Indefinite Pronoun | autres | | PINDFS | Singular Feminine Indefinite Pronoun | chacune | | PINDFP | Plurial Feminine Indefinite Pronoun | certaines | | PROPN | Proper noun | houston | | XFAMIL | Last name | levy | | NUM | Numerical Adjectives | trentaine vingtaine | | DINTMS | Masculine Numerical Adjectives | un | | DINTFS | Feminine Numerical Adjectives | une | | PPOBJMS | Singular Masculine Pronoun complements of objects | le lui | | PPOBJMP | Plurial Masculine Pronoun complements of objects | eux y | | PPOBJFS | Singular Feminine Pronoun complements of objects | moi la | | PPOBJFP | Plurial Feminine Pronoun complements of objects | en y | | PPER1S | Personal Pronoun First Person Singular | je | | PPER2S | Personal Pronoun Second Person Singular | tu | | PPER3MS | Personal Pronoun Third Person Masculine Singular | il | | PPER3MP | Personal Pronoun Third Person Masculine Plurial | ils | | PPER3FS | Personal Pronoun Third Person Feminine Singular | elle | | PPER3FP | Personal Pronoun Third Person Feminine Plurial | elles | | PREFS | Reflexive Pronouns First Person of Singular | me m' | | PREF | Reflexive Pronouns Third Person of Singular | se s' | | PREFP | Reflexive Pronouns First / Second Person of Plurial | nous vous | | VERB | Verb | obtient | | VPPMS | Singular Masculine Participle Past Verb | formulé | | VPPMP | Plurial Masculine Participle Past Verb | classés | | VPPFS | Singular Feminine Participle Past Verb | appelée | | VPPFP | Plurial Feminine Participle Past Verb | sanctionnées | | DET | Determinant | les l' | | DETMS | Singular Masculine Determinant | les | | DETFS | Singular Feminine Determinant | la | | ADJ | Adjective | capable sérieux | | ADJMS | Singular Masculine Adjective | grand important | | ADJMP | Plurial Masculine Adjective | grands petits | | ADJFS | Singular Feminine Adjective | française petite | | ADJFP | Plurial Feminine Adjective | légères petites | | NOUN | Noun | temps | | NMS | Singular Masculine Noun | drapeau | | NMP | Plurial Masculine Noun | journalistes | | NFS | Singular Feminine Noun | tête | | NFP | Plurial Feminine Noun | ondes | | PREL | Relative Pronoun | qui dont | | PRELMS | Singular Masculine Relative Pronoun | lequel | | PRELMP | Plurial Masculine Relative Pronoun | lesquels | | PRELFS | Singular Feminine Relative Pronoun | laquelle | | PRELFP | Plurial Feminine Relative Pronoun | lesquelles | | INTJ | Interjection | merci bref | | CHIF | Numbers | 1979 10 | | SYM | Symbol | € % | | YPFOR | Endpoint | . | | PUNCT | Ponctuation | : , | | MOTINC | Unknown words | Technology Lady | | X | Typos & others | sfeir 3D statu | ## Evaluation results ```plain Results: - F-score (micro): 0.952 - F-score (macro): 0.8644 - Accuracy (incl. no class): 0.952 By class: precision recall f1-score support PPER1S 0.9767 1.0000 0.9882 42 VERB 0.9823 0.9537 0.9678 583 COSUB 0.9344 0.8906 0.9120 128 PUNCT 0.9878 0.9688 0.9782 833 PREP 0.9767 0.9879 0.9822 1483 PDEMMS 0.9583 0.9200 0.9388 75 COCO 0.9839 1.0000 0.9919 245 DET 0.9679 0.9814 0.9746 645 NMP 0.9521 0.9115 0.9313 305 ADJMP 0.8352 0.9268 0.8786 82 PREL 0.9324 0.9857 0.9583 70 PREFP 0.9767 0.9545 0.9655 44 AUX 0.9537 0.9859 0.9695 355 ADV 0.9440 0.9365 0.9402 504 VPPMP 0.8667 1.0000 0.9286 26 DINTMS 0.9919 1.0000 0.9959 122 ADJMS 0.9020 0.9057 0.9039 244 NMS 0.9226 0.9336 0.9281 753 NFS 0.9347 0.9714 0.9527 560 YPFOR 0.9806 1.0000 0.9902 353 PINDMS 1.0000 0.9091 0.9524 44 NOUN 0.8400 0.5385 0.6562 39 PROPN 0.8605 0.8278 0.8439 395 DETMS 0.9972 0.9972 0.9972 362 PPER3MS 0.9341 0.9770 0.9551 87 VPPMS 0.8994 0.9682 0.9325 157 DETFS 1.0000 1.0000 1.0000 240 ADJFS 0.9266 0.9011 0.9136 182 ADJFP 0.9726 0.9342 0.9530 76 NFP 0.9463 0.9749 0.9604 199 VPPFS 0.8000 0.9000 0.8471 40 CHIF 0.9543 0.9414 0.9478 222 XFAMIL 0.9346 0.8696 0.9009 115 PPER3MP 0.9474 0.9000 0.9231 20 PPOBJMS 0.8800 0.9362 0.9072 47 PREF 0.8889 0.9231 0.9057 52 PPOBJMP 1.0000 0.6000 0.7500 10 SYM 0.9706 0.8684 0.9167 38 DINTFS 0.9683 1.0000 0.9839 61 PDEMFS 1.0000 0.8966 0.9455 29 PPER3FS 1.0000 0.9444 0.9714 18 VPPFP 0.9500 1.0000 0.9744 19 PRON 0.9200 0.7419 0.8214 31 PPOBJFS 0.8333 0.8333 0.8333 6 PART 0.8000 1.0000 0.8889 4 PPER3FP 1.0000 1.0000 1.0000 2 MOTINC 0.3571 0.3333 0.3448 15 PDEMMP 1.0000 0.6667 0.8000 3 INTJ 0.4000 0.6667 0.5000 6 PREFS 1.0000 0.5000 0.6667 10 ADJ 0.7917 0.8636 0.8261 22 PINDMP 0.0000 0.0000 0.0000 1 PINDFS 1.0000 1.0000 1.0000 1 NUM 1.0000 0.3333 0.5000 3 PPER2S 1.0000 1.0000 1.0000 2 PPOBJFP 1.0000 0.5000 0.6667 2 PDEMFP 1.0000 0.6667 0.8000 3 X 0.0000 0.0000 0.0000 1 PRELMS 1.0000 1.0000 1.0000 2 PINDFP 1.0000 1.0000 1.0000 1 accuracy 0.9520 10019 macro avg 0.8956 0.8521 0.8644 10019 weighted avg 0.9524 0.9520 0.9515 10019 ``` ## BibTeX Citations Please cite the following paper when using this model. UD_French-GSD corpora: ```latex @misc{ universaldependencies, title={UniversalDependencies/UD_French-GSD}, url={https://github.com/UniversalDependencies/UD_French-GSD}, journal={GitHub}, author={UniversalDependencies} } ``` LIA TAGG: ```latex @techreport{LIA_TAGG, author = {Frédéric Béchet}, title = {LIA_TAGG: a statistical POS tagger + syntactic bracketer}, institution = {Aix-Marseille University & CNRS}, year = {2001} } ``` Flair Embeddings: ```latex @inproceedings{akbik2018coling, title={Contextual String Embeddings for Sequence Labeling}, author={Akbik, Alan and Blythe, Duncan and Vollgraf, Roland}, booktitle = {{COLING} 2018, 27th International Conference on Computational Linguistics}, pages = {1638--1649}, year = {2018} } ```