La dernière fois, vous et moi avons appris à faire des livres parallèles et avons fait une version russe-anglais d'un extrait du roman To Kill a Mockingbird de Harper Lee. Aujourd'hui, nous allons passer à l'étape suivante et créer un livre multilingue complet en huit langues.
Permettez-moi de vous rappeler que l'objectif de ce projet est de créer un outil qui aidera les personnes qui étudient et enseignent des langues étrangères à créer du matériel d'étude et des livres parallèles pour améliorer leurs compétences en lecture. Le problème global est qu'il est difficile de trouver du matériel d'étude intéressant avec une traduction parallèle, alors qu'il est beaucoup plus facile de trouver un livre dans l'original et son édition russe séparément.
Faisons d'abord sept livres
" ", . ( ); , ( ); . , . . pdf .
-.
1.
( , ).
- ( , , , ).
- .
- (H1 , H5 ), .
- , , ( ).
, . , . .
, :
| %%%%%title. | ||
| %%%%%author. | ||
| %%%%%h1. %%%%%h2. %%%%%h3. %%%%%h4. %%%%%h5. | ||
| %%%%%divider. | ||
| %%%%%. |
, , . .
, :
%%%%%title. .. %%%%%author. %%%%%h1. 1%%%%%h2. %%%%%h2. , , , , . , , , , , , . – , , – , . ...
MIHAIL BULGAKOV%%%%%author. A MESTER ÉS MARGARITA%%%%%title. ELSŐ KÖNYV%%%%%h1. ELSŐ FEJEZET%%%%%h2. Ne álljunk szóba ismeretlenekkel%%%%%h2. Egy meleg tavaszi estén, az alkonyat órájában, a Patriarsije Prudin két férfiú jelent meg. Az egyik negyvenéves forma, kövérkés, alacsony, kopasz fekete emberke, szürke nyári öltönyt viselt, elegáns kalapját kezében tartotta, gondosan borotvált arcát istentelen nagy méretű, fekete csontkeretes pápaszem ékesítette. Társa jóval fiatalabb s vállasabb volt nála, borzas haján tarkójáig hátratolt kockás sapka; öltözéke kockás sportingből, gyűrött fehér nadrágból, fekete szandálból állt. ...
. - , - . , .
2.
Colab
Colab , . . html .
pip install lingtrain-aligner
:
# from lingtrain_aligner import preprocessor, splitter, aligner, resolver, reader, vis_helper # text1_input = "master_hu.txt" text2_input = "master_ru.txt" with open(text1_input, "r", encoding="utf8") as input1: text1 = input1.readlines() with open(text2_input, "r", encoding="utf8") as input2: text2 = input2.readlines() # , db_path = "master_hu.db" # ( ) lang_from = "hu" lang_to = "ru" # models = ["sentence_transformer_multilingual", "sentence_transformer_multilingual_labse"] model_name = models[0] # text1_prepared = preprocessor.mark_paragraphs(lines1_prepared) text2_prepared = preprocessor.mark_paragraphs(lines2_prepared) # splitted_from = splitter.split_by_sentences_wrapper(lines1_prepared, lang_from, leave_marks=True) splitted_to = splitter.split_by_sentences_wrapper(lines2_prepared, lang_to, leave_marks=True)
9460 8996 . - , .
. SQLite , , .
aligner.fill_db(db_path, lang_from, lang_to, splitted_from, splitted_to)
, :
batch_ids = range(10) aligner.align_db(db_path, model_name, batch_size=100, window=60, batch_ids=batch_ids, save_pic=False, embed_batch_size=25, normalize_embeddings=True, show_progress_bar=True )
— ( ), 10 (batch_ids — ) 100 ( batch_size). window "" , . , " " .
, . ( ), shift ( ).
vis_helper:
vis_helper.visualize_alignment_by_db(db_path, output_path="alignment_vis.png", batch_size=100, size=(800,800), lang_name_from=lang_from, lang_name_to=lang_to, batch_ids=batch_ids, plt_show=True)
batch_ids , . , .
, . , " " , .
.
resolver. , , . , , . , . , — :
steps = 3 batch_id = -1 # for i in range(steps): # conflicts, rest = resolver.get_all_conflicts(db_path, min_chain_length=2+i, max_conflicts_len=6*(i+1), batch_id=batch_id) # resolver.get_statistics(conflicts) resolver.get_statistics(rest) # resolver.resolve_all_conflicts(db_path, conflicts, model_name, show_logs=False) if len(rest) == 0: break
. , :
# conflicts, rest = resolver.get_all_conflicts(db_path, min_chain_length=2, max_conflicts_len=20, batch_id=-1) # resolver.get_statistics(conflicts) resolver.get_statistics(rest) # resolver.resolve_all_conflicts(db_path, conflicts, model_name, show_logs=False) # vis_helper.visualize_alignment_by_db(db_path, output_path="alignment_vis.png", batch_size=100, size=(800,800), lang_name_from=lang_from, lang_name_to=lang_to, batch_ids=batch_ids, plt_show=False)
. , . 4000 . 10 100 , :
batch_ids = range(10, 50) aligner.align_db(db_path, \ model_name, \ batch_size=100, \ window=60, \ batch_ids=batch_ids, \ save_pic=False, embed_batch_size=25, \ normalize_embeddings=True, \ show_progress_bar=True )
( , ) , . , . 9460, , 100 95 .
batch_ids = range(50, 100)
( ) . master_hu, - " ".
3.
, reader, - :
from lingtrain_aligner import reader # paragraphs_from, paragraphs_to, meta = reader.get_paragraphs(db_path, direction="from") # html reader.create_book(paragraphs_from, paragraphs_to, meta, output_path = "lingtrain.html")
html . . , :
reader.create_book(paragraphs_from, paragraphs_to, meta, output_path = f"lingtrain.html", template="pastel_fill")
, :
# span' my_style = [ '{"background": "linear-gradient(90deg, #FDEB71 0px, #fff 150px)", "border-radius": "15px"}', '{"background": "linear-gradient(90deg, #ABDCFF 0px, #fff 150px)", "border-radius": "15px"}', '{"background": "linear-gradient(90deg, #FEB692 0px, #fff 150px)", "border-radius": "15px"}', '{"background": "linear-gradient(90deg, #CE9FFC 0px, #fff 150px)", "border-radius": "15px"}', '{"background": "linear-gradient(90deg, #81FBB8 0px, #fff 150px)", "border-radius": "15px"}' ] reader.create_book(paragraphs_from, paragraphs_to, meta, output_path = f"lingtrain.html", template="custom", styles=my_style)
, , master_hu.db . , (, , , ).
, .
1. – ! – , – , , ! 2. - - , – , , .
1. 'Pah, the devil!' exclaimed the editor. 2. 'You know, Ivan, I nearly had heat stroke just now! 3. There was even something like a hallucination...' 4. He attempted to smile, but alarm still jumped in his eyes and his hands trembled.
9707 8996 . , window=100 1000 .
, .
, .
, , . (8190 ) . 100 , — :
batch_ids = range(10,12) aligner.align_db(db_path, \ model_name, \ batch_size=200, \ window=100, \ batch_ids=batch_ids, \ save_pic=False, embed_batch_size=25, \ normalize_embeddings=True, \ show_progress_bar=True )
, shift. :
batch_ids = range(10,12) aligner.align_db(db_path, \ model_name, \ batch_size=200, \ window=100, \ batch_ids=batch_ids, \ save_pic=False, embed_batch_size=25, \ normalize_embeddings=True, \ show_progress_bar=True, shift=-100 # )
, , . .
:
, - :
, - . - ( ), , , . 20 , .
master_*.
, (). , .
:
, ( ) . , . , . , ( *.db ).
, . , .
:
par_struct_1 = [2, 5, 6, 8, 10, 12, ...] par_struct_2 = [2, 6, 8, 10, 11, 12, ...] res = [2, 6, 8, 10, 12, ...]
, :
, .
, , :
, . , , — .
. :
[ -- [ -- [1, "[1,2]", 1, "[1]"], -- .. [2, "[3]", 2, "[2]"], [3, "[4,5]", 3, "[3]"], [4, "[6]", 4, "[4,5,6]"], ... ], -- [ [1, "[100]", 1, "[105]"], ... ], ... ]
"" "". — , — , .
:
- ,
- ,
- ,
, . , . :
[[1], [2], [3]] -> [[1, 2, 3], [4], [5]] -> [[1], [2], [3]] [[4]] -> [[6]] -> [[4]] [[5]] -> [[7]] -> [[5, 6]] ...
— , — . .
, :
( ), , . , :
4.
get_paragraphs() create_book(). , .
get_paragraphs_polybook(), . , , . ( ):
from lingtrain_aligner import resolver conflicts_to_solve, rest = resolver.get_all_conflicts(db_path, min_chain_length=2, max_conflicts_len=20, batch_id=-1) # resolver.get_statistics(conflicts_to_solve) resolver.get_statistics(rest)
:
from lingtrain_aligner import reader db_path1 = "db/master_de.db" db_path2 = "db/master_uk.db" db_path3 = "db/master_be.db" db_path4 = "db/master_zh.db" db_path5 = "db/master_cz.db" db_path6 = "db/master_en.db" db_path7 = "db/master_hu.db" paragraphs, metas = reader.get_paragraphs_polybook( db_paths=[db_path1, db_path2, db_path3, db_path4, db_path5, db_path6, db_path7])
Maintenant, la chose la plus intéressante est que nous pouvons choisir les langues dont nous avons besoin et créer notre livre. Vous pouvez, par exemple, prendre l'anglais et l'allemand, beaucoup apprennent ces langues à deux.
reader.create_polybook( lang_ordered=["en", "de", "ru"], paragraphs = paragraphs, delimeters = delimeters, metas = metas, output_path = "lingtrain_master.tml", template="none")
Le paramètre lang_ordered spécifie l'ensemble et l'ordre des langues.
Vous pouvez faire une version biélorusse-chinoise :
reader.create_polybook( lang_ordered=["be", "zh"], paragraphs = paragraphs, delimeters = delimeters, metas = metas, output_path = "lingtrain_master.html", template="none")
Vous pouvez bien sûr tout prendre :
reader.create_polybook( lang_ordered = ["ru", "en", "de", "be", "uk", "cz", "hu", "zh"], paragraphs = paragraphs, delimeters = delimeters, metas = metas, output_path = "lingtrain_master_total.html", template="none")
Il est préférable de laisser ces alignements (fichiers * .db), dont vous avez besoin des langues. Cela gardera autant d'informations sur les paragraphes que possible. Avec les styles, tout est comme avant, vous pouvez définir un modèle (maintenant il y a "pastel_fill", "pastel_start"), vous pouvez inventer le vôtre - les exemples sont ci-dessus et dans le cahier sur Colab. Si vous définissez template = "none" , vous obtenez un livre vierge.
Plans et soutien
- fb2 . , . -, .
- , http://t.me/lingtrain_books.
- . - .
:
[3] Google Colab
[5] PDF. Sept livres parallèles "Le Maître et Marguerite"