Начало >
Доклад за качеството на превода на Lingvanex

Доклад за качеството на превода на Lingvanex

October 22, 2024

Целта на този доклад е да покаже качеството на превода на езиковите модели на Lingvanex според две най-популярни метрики за оценка на машинен превод.

Flores е набор от тестове с отворен код и публично достъпни данни, пуснат от Facebook Research и има най-голямото покритие на езиковите двойки.

Описание на показателите за качество

BLEU

BLEU е автоматична метрика, базирана на n-грамове. Той измерва прецизността на n-грамове на изхода от машинен превод в сравнение с референтния, претеглен с наказание за краткост, за да накаже твърде кратки преводи. Ние използваме специфична реализация на BLEU, наречена sacreBLEU. Той извежда резултати от корпуса, а не от сегменти.

Референции

Papineni, Kishore, S. Roukos, T. Ward and Wei-Jing Zhu. “Bleu: a Method for Automatic Evaluation of Machine Translation.” ACL (2002).
Post, Matt. “A Call for Clarity in Reporting BLEU Scores.” WMT (2018).

COMET

COMET (Crosslingual Optimized Metric for Evaluation of Translation) е метрика за автоматична оценка на машинен превод, която изчислява приликата между изход от машинен превод и референтен превод с помощта на вграждане на токени или изречения. За разлика от други показатели, COMET е обучен да предсказва различни видове човешки преценки под формата на усилие след редактиране, директна оценка или анализ на грешки в превода.

Референции

COMET - https://machinetranslate.org/comet
COMET: High-quality Machine Translation Evaluation - https://unbabel.github.io/COMET/html/index.html#comet-high-quality-machine-translation-evaluation

Езикови двойки

Забележка: По-малкият размер на моделите на твърдия диск означава по-ниска консумация на GPU памет, което води до намалени разходи за внедряване. По-малкият размер на модела има по-добра производителност във времето за превод. Приблизителното използване на GPU памет се изчислява като размер на модела на твърдия диск x 1,2

Езикова двойка	На модела Размер, mb	Тест данни	BLEU Метрика	COMET Метрика
Afrikaans - English	113,91	Flores	52,33	86,43
English - Afrikaans	113,91	Flores	42,52	86,92
Albanian - English	113,91	Lingvanex	56,43	87,83
English - Albanian	113,91	Lingvanex	57,04	89,03
Amharic - English	113,91	Flores	30,42	83,78
English - Amharic	113,91	Flores	14,13	87,4
Arabic - English	184,11	Flores	45,79	87,95
English - Arabic	190,63	Flores	33,45	88,28
Armenian - English	113,91	Flores	37,47	86,93
English - Armenian	113,91	Flores	22,09	89,32
Armenian - Russian	190,63	Flores	20,97	85,06
Russian - Armenian	190,63	Flores	15,71	87,00
Azerbaijani - English	113,91	Flores	23,76	85,88
English - Azerbaijani	113,91	Flores	17,79	84,93
Azerbaijani - Russian	190,63	Flores	16,95	85,23
Russian - Azerbaijani	190,63	Flores	13,10	83,64
Basque - English	113,91	Flores	32,34	86,40
English - Basque	113,91	Flores	21,69	86,99
Belarusian - English	113,91	Flores	19,77	80,89
English - Belarusian	113,91	Flores	15,54	85,16
Bengali - English	113,91	Flores	33,93	87,86
English - Bengali	239,56	Flores	23,41	86,92
Bosnian - English	113,91	Flores	41,62	87,65
English - Bosnian	113,91	Flores	35,57	90,61
Bulgarian - English	184,00	Flores	44,97	88,28
English - Bulgarian	184,00	Flores	45,90	91,22
Catalan - English	113,91	Flores	47,54	88,55
English - Catalan	113,91	Flores	45,62	87,67
Cebuano - English	113,91	Flores	41,17	79,24
English - Cebuano	113,91	Flores	35,46	71,80
Chichewa - English	113,91	Flores	22,86	69,22
English - Chichewa	113,91	Flores	17,44	64,71
Chinese (Simplified) - English	184,11	Flores	29,91	86,27
English - Chinese (Simplified)	184,11	Flores	42,29	87,70
Chinese (Traditional) - English	190,63	Flores	30,28	86,51
English - Chinese (Traditional)	190,63	Flores	36,06	88,67
Corsican - English	113,91	Lingvanex	59,90	83,86
English - Corsican	113,91	Lingvanex	54,21	76,34
Croatian - English	184,00	Flores 101	39,80	88,08
English - Croatian	184,00	Flores	34,95	91,09
Czech - English	184,00	Flores	41,83	88,33
English - Czech	184,00	Flores	36,53	90,98
Danish - English	184,00	Flores	51,21	90,25
English - Danish	184,02	Flores	49,88	91,27
Dutch - English	184,00	Flores	34,04	87,20
English - Dutch	184,00	Flores	29,71	88,10
Esperanto - English	113,91	Flores	41,60	87,49
English - Esperanto	113,91	Flores	32,02	88,92
Estonian - English	113,91	Flores	41,07	89,33
English - Estonian	113,91	Flores	31,48	91,64
Filipino - English	113,91	Flores	47,41	86,85
English - Filipino	113,91	Flores	37,75	84,61
Finnish - English	184,00	Flores	36,87	89,76
English - Finnish	184,00	Flores	27,06	90,96
French - English	190,65	Flores	48,82	89,46
English - French	190,65	Flores	53,25	88,40
Frisian - English	113,91	Lingvanex	65,06	84,99
English - Frisian	113,91	Lingvanex	54,82	81,09
Gaelic-English	113,91	Flores	29,60	76,02
English-Gaelic	113,91	Flores	24,72	80,92
Galician - English	113,91	Flores	40,58	87,52
English - Galician	113,91	Flores	37,71	87,03
Georgian - Russian	190,63	Flores	18,60	85,80
Russian - Georgian	190,63	Flores	13,85	85,77
Georgian - English	113,91	Flores	22,09	89,32
English - Georgian	113,91	Flores	16,56	87,13
German - English	190,65	Flores	46,76	89,34
English - German	184,02	Flores	43,09	88,15
Greek - English	184,00	Flores	36,19	87,38
English - Greek	184,00	Flores	29,54	89,02
Gujarati-English	113,91	Flores	142,09	89,68
English - Gujarati	113,91	Flores	28,00	88,61
Haitian Creole - English	113,91	Flores	30,08	72,24
English - Haitian Creole	113,91	Flores	28,27	67,41
Hausa - English	113,91	Flores	33,64	78,80
English - Hausa	113,91	Flores	28,70	80,76
Hawaiian - English	113,91	Lingvanex	45,12	71,92
English - Hawaiian	113,91	Lingvanex	64,27	78,56
Hebrew - English	184,11	Flores	46,31	88,32
English - Hebrew	184,11	Flores	36,00	88,53
Hindi - English	113,91	Flores	38,18	88,29
English - Hindi	113,91	Flores	67,62	81,99
English - Hmong	113,91	Lingvanex	60,99	77,35
Hmong - English	184,00	Lingvanex	45,87	75,22
Hungarian - English	184,00	Flores	38,74	88,37
English - Hungarian	184,00	Flores	31,40	89,97
Icelandic - English	113,91	Flores	37,27	85,92
English - Icelandic	113,91	Flores	31,90	86,34
Igbo - English	113,91	Flores	26,60	68,85
English - Igbo	113,91	Flores	19,01	71,94
Indonesian - English	184,02	Flores	46,65	89,60
English - Indonesian	184,02	Flores	50,77	91,90
Italian - English	184,00	Flores	34,74	87,89
English - Italian	184,00	Flores	32,57	88,19
Irish - English	113,91	Flores	44,15	85,29
English - Irish	113,91	Flores	38,86	81,81
Japanese - English	190,63	Flores	31,05	88,08
English - Japanese	190,63	Flores	39,62	91,56
Javanese - English	113,91	Flores	30,29	76,20
English - Javanese	113,91	Flores	28,79	86,59
Kannada - English	113,91	Flores	36,01	87,38
English - Kannada	113,91	Flores	65,83	86,33
Kazakh - English	113,91	Flores	36,25	87,61
English - Kazakh	113,91	Flores	26,48	90,35
Kazakh - Russian	190,63	Flores	22,79	88,12
Russian - Kazakh	190,63	Flores	19,18	89,57
Khmer - English	113,91	Flores	33,19	85,77
English - Khmer	113,91	Flores	4,77	82,37
Kinyarwanda - English	113,91	Flores	32,98	73,73
English - Kinyarwanda	113,91	Flores	25,84	66,67
Korean - English	113,91	Flores	32,85	88,09
English - Korean	113,91	Flores	33,66	89,67
Kurdish - English	113,91	Flores	28,81	79,90
English - Kurdish	113,91	Flores	12,85	80,84
Kyrgyz - English	113,91	Flores	24,29	84,79
English - Kyrgyz	113,91	Flores	16,65	88,14
Kyrgyz - Russian	190,63	Flores	16,38	86,32
Russian - Kyrgyz	190,63	Flores	13,43	87,62
Lao - English	113,91	Flores	31,45	84,11
English - Lao	113,91	Flores	59,54	83,24
Latin - English	113,91	Lingvanex	24,24	74,81
English - Latin	113,91	Lingvanex	14,83	77,64
Latvian - English	113,91	Flores	38,95	87,67
English - Latvian	113,91	Flores	37,88	90,49
Lithuanian - English	113,91	Flores	34,96	86,24
English - Lithuanian	113,91	Flores	31,28	90,11
Luxembourgish - English	113,91	Flores	45,56	80,29
English - Luxembourgish	113,91	Flores	27,28	57,29
Macedonian - English	113,91	Flores	44,33	87,43
English - Macedonian	113,91	Flores	38,19	89,28
Malagasy - English	113,91	Lingvanex	38,60	83,10
English - Malagasy	113,91	Lingvanex	41,74	85,25
Malay - English	184,11	Flores	46,09	88,84
English - Malay	184,11	Flores	44,63	89,77
Malayalam - English	113,91	Flores	39,26	88,28
English - Malayalam	113,91	Flores	23,19	88,92
Maltese - English	113,91	Flores	53,11	81,83
English - Maltese	113,91	Flores	47,57	79,52
Maori - English	113,91	Flores	29,13	69,87
English - Maori	113,91	Flores	17,44	61,51
Marathi - English	113,91	Flores	37,94	87,49
English - Marathi	113,91	Flores	21,42	75,73
Mongolian- English	113,91	Flores	31,40	85,43
English - Mongolian	113,91	Flores	20,07	89,00
Myanmar - English	113,91	Flores	23,13	82,43
English - Myanmar	113,91	Flores	55,72	87,25
Nepali - English	113,91	Flores	41,67	89,94
English - Nepali	113,91	Flores	64,92	83,27
Norwegian - English	184,00	Flores	44,20	88,78
English - Norwegian	184,02	Flores	35,85	90,25
Odia - English	113,91	Flores	32,34	86,77
English - Odia	113,91	Flores	21,09	81,84
Persian - English	184,02	Flores	39,04	87,71
English - Persian	184,11	Flores	27,24	87,82
Polish - English	184,00	Flores	30,94	85,72
English - Polish	184,00	Flores	24,24	89,21
Portuguese (Brazil) - English	184,02	Flores	51,42	89,66
English - Portuguese (Brazil)	184,02	Flores	51,67	89,40
Portuguese - English	184,02	Flores	50,93	89,27
English - Portuguese	184,02	Flores	51,38	89,71
Punjabi - English	113,91	Flores	38,15	87,74
English - Punjabi	113,91	Flores	28,91	84,28
Pushto - English	113,91	Flores	27,05	79,68
English - Pushto	113,91	Flores	15,04	77,94
Romanian - English	184,00	Flores	45,22	89,30
English - Romanian	179,58	Flores	43,68	90,29
Russian - English	190,65	Flores	37,86	86,63
English - Russian	190,65	Flores	34,64	89,37
Russian - Belarusian	190,63	Lingvanex	57,35	95,38
Samoan - English	113,91	Flores	27,81	68,62
English - Samoan	113,91	Flores	30,07	67,20
Serbian - English	184,00	Flores	43,06	86,68
English - Serbian	184,00	Flores	36,87	88,49
Sesotho - English	113,91	Flores	32,91	72,64
English - Sesotho	113,91	Flores	20,36	66,18
Shona - English	113,91	Flores	23,05	70,04
English - Shona	113,91	Flores	13,74	62,88
Sindhi - English	113,91	Flores	31,09	79,66
English - Sindhi	113,91	Flores	25,90	80,37
Sinhala - English	113,91	Flores	36,73	87,92
English - Sinhala	113,91	Flores	21,42	87,65
Slovak - English	184,11	Flores	42,32	88,33
English - Slovak	184,11	Flores	40,08	91,01
Slovenian - English	113,91	Flores	38,25	87,76
English - Slovenian	113,91	Flores	35,39	90,20
Somali - English	113,91	Flores	29,89	77,25
English - Somali	113,91	Flores	14,56	80,85
Spanish - English	184,02	Flores	31,48	86,92
English - Spanish	184,02	Flores	30,17	86,72
Sundanese - English	113,91	Flores	34,02	81,42
English - Sundanese	113,91	Flores	20,23	79,65
Swahili - English	113,91	Flores	44,95	84,91
English - Swahili	113,91	Flores	41,14	85,40
Swedish - English	184,00	Flores	51,89	90,13
English - Swedish	184,00	Flores	49,39	91,24
Tajik - English	113,91	Flores	33,74	77,46
English - Tajik	113,91	Flores	24,97	76,28
Tajik - Russian	190,63	Flores	23,15	79,69
Russian - Tajik	190,63	Flores	19,38	75,59
Tamil - English	113,91	Flores	35,17	86,48
English - Tamil	113,91	Flores	23,12	89,55
Tatar - English	113,91	Flores	30,37	79,43
English - Tatar	113,91	Flores	19,01	69,26
Telugu-English	113,91	Flores	42,02	87,74
English-Telugu	113,91	Flores	28,30	87,60
Thai - English	184,02	Flores	31,63	87,71
English - Thai	113,91	Flores	63,68	88,57
Turkish - English	184,00	Flores	41,85	89,63
English - Turkish	184,00	Flores	35,33	90,82
Turkmen - English	113,91	Flores	37,10	80,90
English - Turkmen	113,91	Flores	22,91	67,00
Ukrainian - English	184,00	Flores	41,54	86,98
English - Ukrainian	184,00	Flores	34,30	89,88
Ukrainian - Russian	190,63	Flores	28,07	91,66
Russian - Ukrainian	190,63	Flores	25,62	92,16
Urdu - English	113,91	Flores	31,09	84,46
English - Urdu	113,91	Flores	24,80	81,71
Uyghur - English	113,91	Flores	28,17	86,26
English - Uyghur	113,91	Flores	22,66	86,21
Uzbek - English	113,91	Flores	37,42	87,64
English - Uzbek	113,91	Flores	23,04	90,32
Uzbek - Russian	190,63	Flores	22,11	85,40
Russian - Uzbek	190,63	Flores	16,86	87,37
Vietnamese - English	184,00	Flores	38,65	87,53
English - Vietnamese	184,11	Flores	46,28	89,16
Welsh - English	113,91	Flores	61,36	89,32
English - Welsh	113,91	Flores	57,55	88,83
Xhosa - English	113,91	Flores	36,65	78,10
English - Xhosa	113,91	Flores	18,90	77,05
Yiddish - English	113,91	Flores	41,89	76,88
English - Yiddish	113,91	Flores	10,88	67,67
Yoruba - English	113,91	Flores	17,54	61,56
English - Yoruba	113,91	Flores	3,57	55,39
Zulu - English	113,91	Flores	36,83	78,00
English - Zulu	113,91	Flores	21,06	78,12

Доклад за качеството на превода на Lingvanex

Описание на показателите за качество

BLEU

Референции

COMET

Референции

Езикови двойки

Поискайте повече информация

Завършено