Dataset Viewer
Auto-converted to Parquet Duplicate
sample_id
large_stringlengths
10
10
speaker_id
large_stringclasses
2 values
reference_audio_path
audioduration (s)
1.68
20.2
reference_duration_s
float64
1.68
20.2
reference_text_en
large_stringlengths
19
274
target_audio_path
audioduration (s)
1.77
14.5
target_duration_s
float64
1.76
14.5
target_text_en
large_stringlengths
39
196
target_text_ar
large_stringlengths
25
174
target_text_fr
large_stringlengths
37
311
train_0000
speaker_000
3.51
For instance, in Spanish we use words that come from English.
1.76
Um. Well, this is everything that I have.
ممم. حسنًا، هذا كل ما لدي.
Hum. Eh bien, c’est tout ce que j’ai.
train_0001
speaker_000
3.39
There are however some differences between lexical borrowing and code-switching.
4.31
In fact, ninety two percent of the borrowings in the test set are OOV.
وفي الواقع، اثنان وتسعون في المئة من الاقتراضات في مجموعة الاختبار هي OOV.
En effet, quatre-vingt-douze pour cent des emprunts dans l’ensemble de test sont des OOV.
train_0002
speaker_000
10.28
So, this proves that the dataset that we created is more difficult and that we needed to explore more sophisticated models for these tasks.
4.47
We have explored four types of models for lexical borrowing detection.
لقد استكشفنا أربعة أنواع من النماذج للكشف عن الاقتراض المعجمي.
Nous avons exploré quatre types de modèles pour la détection d’emprunt lexical.
train_0003
speaker_000
1.72
So, they were not seen during training.
7.9
One for English lexical borrowings which is the majority of lexical borrowings in Spanish, and then the label other for borrowings from other languages.
واحدة ل الاقتراضات الإنجليزية المعجمية التي هي غالبية الاقتراضات المعجمية في اللغة الإسبانية، والعلامة الأخرى ل الاقتراضات من لغات أخرى.
Une pour les emprunts lexicaux anglais, constituant la majorité des emprunts lexicaux en espagnol, puis l’autre étiquette pour les emprunts d’autres langues.
train_0004
speaker_000
4.31
In fact, ninety two percent of the borrowings in the test set are OOV.
10.83
Just ah to bear in mind that the best result that we got with multilingual BERT obtained an F1 of seventy six on the development set and eighty two on the test set.
يجب أن نضع في اعتبارنا أن أفضل نتيجة حصلنا عليها مع بيرت متعدد اللغات كانت ستة وسبعين درجة F1 في مجموعة التطوير واثنين وثمانين في مجموعة الاختبار.
Gardons juste à l’esprit que le meilleur résultat que nous avons obtenu avec Représentations d'encodeurs bidirectionnels à partir de transformateurs multilingues était un F1 de soixante-seize sur l’ensemble de développement et quatre-vingt-deux sur l’ensemble de test.
train_0005
speaker_000
7.06
Well, from the point of view of linguistics, borrowing is a manifestation of of how languages change and how they interact.
7.58
So the BIO was used for words in Spanish so not for words that were not borrowed.
إذًا فقد تم استخدام BIO ل الكلمات باللغة الإسبانية وليس ل الكلمات غير المقترضة.
Donc le BIO a été utilisé pour des mots en espagnol, et non pour des mots qui n’ont pas été empruntés.
train_0006
speaker_000
10.96
Ah the aim at a new dataset that was annotated with lexical borrowings and the aim was to create a test set that was as difficult as possible.
8.84
As you can see, multilingual BERT performs better than BETO both on the development set and on the test set and across all metrics.
كما ترى، أداء بيرت متعدد اللغات أفضل من BETO في مجموعة التطوير وفي مجموعة الاختبار وعبر جميع المقاييس.
Comme vous pouvez le voir, les Représentations d'encodeurs bidirectionnels à partir de transformateurs multilingues fonctionnent mieux que les BETO à la fois sur l’ensemble de développement et sur l’ensemble de test, et à travers tous les indicateurs.
train_0007
speaker_000
4.59
But to begin with, what is lexical borrowing and why it matters as an NLP task?
7.8
And we tried experimented with different type of embeddings like transformer-based but also fast-text, character embeddings, and so on.
وأجرينا تجارب باستخدام أنواع مختلفة من التضمينات مثل التضمينات التي تستند إلى محول ولكنها تستند إلى النص السريع كذلك، والتي تستند إلى الحروف، وما إلى ذلك.
Et nous avons essayé d’expérimenter différents types d’intégrations comme celles basées sur la conversion, mais aussi des intégrations de caractères, de texte rapide et ainsi de suite.
train_0008
speaker_000
5.5
Well, lexical borrowing is basically the incorporation of words from one language into another language.
7.9
One for English lexical borrowings which is the majority of lexical borrowings in Spanish, and then the label other for borrowings from other languages.
واحدة ل الاقتراضات الإنجليزية المعجمية التي هي غالبية الاقتراضات المعجمية في اللغة الإسبانية، والعلامة الأخرى ل الاقتراضات من لغات أخرى.
Une pour les emprunts lexicaux anglais, constituant la majorité des emprunts lexicaux en espagnol, puis l’autre étiquette pour les emprunts d’autres langues.
train_0009
speaker_000
6.7
This is a sentence in Spanish: Las prendas bestsellers se estampan con motivos florales, animal print o retales tipo patchwork.
8.5
And here you have the number of spans that were labeled as English and the spans that were labeled as other borrowings and how many of them were unique.
وهنا لديك عدد الامتدادات التي تم تصنيفها على أنها إنجليزية و الامتدادات التي تم تصنيفها على أنها اقتراضات أخرى وكم منها كانت فريدة من نوعها.
Et ici, vous avez le nombre d’étendages qui ont été étiquetés comme anglais et les étendages qui ont été étiquetés comme autres emprunts, et combien d’entre eux étaient uniques.
train_0010
speaker_000
3.89
The borrowings will comply with the grammar of the recipient language.
12.37
We use CONLL formats and we used BIO encoding so that we could encode ah single token borrowings such as app or multi token borrowings such as machine learning.
نستخدم تنسيقات CONLL واستخدمنا ترميز BIO حتى نتمكن من ترميز الاقتراضات أحادية الرمز مثل app أو الاقتراضات متعددة الرموز مثل machine learning.
Nous utilisons les formats CoNLL et avons utilisé l’encodage BIO pour pouvoir encoder des emprunts de gages simples tels que app ou des emprunts de gages multiples tels que l'apprentissage automatique.
train_0011
speaker_000
5.11
In terms of NLP ah borrowings are a common source of out-of-vocabulary words.
5.5
LinCE is a dataset on code switching that has a section on Spanish English, Spanish English code switching.
LinCE هو عبارة عن مجموعة بيانات للتبديل اللغوي بها قسم عن التبديل اللغوي الإسباني الإنجليزي.
LinCE est un ensemble de données sur l’alternance codique qui comporte une section sur l’anglais espagnol, l’alternance codique anglais espagnol.
train_0012
speaker_000
4.89
And we used the same handcrafted features from that from those from that work.
4.73
Um. In terms of error analysis, well, recall was a weak point for all models.
ممم. من حيث تحليل الخطأ؛ حسنًا، كان الاستدعاء نقطة ضعف لكل النماذج.
Hum. En termes d’analyse des erreurs, eh bien, le rappel était un point faible pour tous les modèles.
train_0013
speaker_000
2.24
This model achieved an F1 score of eighty six.
7.8
And we tried experimented with different type of embeddings like transformer-based but also fast-text, character embeddings, and so on.
وأجرينا تجارب باستخدام أنواع مختلفة من التضمينات مثل التضمينات التي تستند إلى محول ولكنها تستند إلى النص السريع كذلك، والتي تستند إلى الحروف، وما إلى ذلك.
Et nous avons essayé d’expérimenter différents types d’intégrations comme celles basées sur la conversion, mais aussi des intégrations de caractères, de texte rapide et ainsi de suite.
train_0014
speaker_000
5.42
So what we aim is to tackle some of these limitations in the task.
4.33
Both models we use them through the transformers library by HuggingFace.
ونستخدم كلا النموذجين من خلال مكتبة المحولات بواسطة HuggingFace.
Nous utilisons les deux modèles à travers la bibliothèque de conversion d’HuggingFace.
train_0015
speaker_000
6.52
And as a result, well, the test set comes from sources and dates that we're not seeing in the training set.
2.23
As you can see, these are the features.
فكما ترى، هذه هي الميزات.
Comme vous pouvez le voir, voici les fonctions.
train_0016
speaker_000
8.84
As you can see, multilingual BERT performs better than BETO both on the development set and on the test set and across all metrics.
5.5
LinCE is a dataset on code switching that has a section on Spanish English, Spanish English code switching.
LinCE هو عبارة عن مجموعة بيانات للتبديل اللغوي بها قسم عن التبديل اللغوي الإسباني الإنجليزي.
LinCE est un ensemble de données sur l’alternance codique qui comporte une section sur l’anglais espagnol, l’alternance codique anglais espagnol.
train_0017
speaker_000
3.74
Ah, this was the model that had been used on previous work.
14.52
So, we run the same BiLSTM-CRF model that we had run using flare, but instead of using these unadapted transformer-based BETO and BERT embeddings, we used code switch embeddings.
لذلك، نقوم بإجرء نفس نموذج BiLSTM-CRF الذي أجريناه باستخدام flare، ولكن بدلاً من استخدام تضمينات BETO وبيرتغير المتكيفة التي تستند إلى محول، استخدمنا تضمينات التبديل اللغوي.
Nous exécutons alors le même modèle BiLSTM-CRF que nous avions exécuté en utilisant Flare, mais au lieu d’utiliser ces intégrations de BETO et Représentations d'encodeurs bidirectionnels à partir de transformateurs basées sur la conversion non adaptées, nous avons utilisé les intégrations d’alternance codique.
train_0018
speaker_000
4.73
And ah it was annotated by hand ah using two tags.
8.93
The CRF model obtained a fifty five obtained a fifty five F1 score, whereas the multilingual BERT obtained eighty two, which is a big difference.
وحصل نموذج CRF على خمسة وخمسين درجة F1، في حين أن بيرت متعدد اللغات حصل على اثنين وثمانين، وهو فرق كبير.
Le modèle CRF a obtenu cinquante-cinq points en F1, tandis que les Représentations d'encodeurs bidirectionnels à partir de transformateurs multilingues ont obtenu quatre-vingt-deux, ce qui est une grande différence.
train_0019
speaker_000
8.35
We obtain fifty five F1 score using the the CRF model with handcrafted features.
2.65
And character embedding seem to improve recall.
ويبدو أن تضمين الحروف يحسن الاستدعاء.
Et l’intégration de caractères semble améliorer le rappel.
train_0020
speaker_001
6.68
As you can see, Kinyarwanda words contains several morphemes that contain different information in them.
11.64
All models except the multilingual are pretrained for thirty two thousand gradient updates with a batch size of two thousand five hundred and sixty sequences in each batch.
جميع النماذج باستثناء النموذج متعدد اللغات مدربة مسبقًا على اثنين وثلاثين ألف تحديث تدرج مع حجم دفعة مكون من ألفين وخمسمائة وستين تسلسل في كل دفعة.
Tous les modèles sauf les multilingues sont préformés pour trente-deux mille mises à jour de pentes avec une taille de lot de deux mille cinq cent soixante séquences dans chaque lot.
train_0021
speaker_001
6.69
We further concat we further concatenate them with another stem embedding at the sentence level.
11.12
We effectively model the morphology of all Kinyarwanda words, including verbals, nouns, demonstrative and possessive pronouns, numerals, and others.
نقوم بنمذجة مورفولوجيا جميع الكلمات الكينيرواندية، بما في ذلك الشفهيات، والأسماء، والضمائر التوضيحية وضمائر الملكية، والأرقام، وغيرها.
Nous modélisons efficacement la morphologie de tous les mots en kinyarwanda, y compris les verbes, les noms, les pronoms démonstratifs et possessifs, les chiffres et autres.
train_0022
speaker_001
4.39
One, the morphology encoder does not use any positional encoding.
12.47
During pre-training, fifteen percent of all words are considered for prediction, of which eighty percent are masked, ten percent are swapped with random words, and ten percent are left unchanged.
وخلال مرحلة ما قبل التدريب، تعتبر خمسة عشر في المئة من جميع الكلمات للتنبؤ، منها ثمانين في المئة مقنعة، ويتم تبديل عشرة في المئة بكلمات عشوائية، وترك عشرة في المئة دون تغيير.
Pendant la préformation, quinze pour cent de tous les mots sont considérés pour la prévention, dont quatre-vingt pour cent sont masqués, dix pour cent sont échangés avec des mots aléatoires et dix pour cent sont laissés inchangés.
train_0023
speaker_001
3.78
Today, I'll talk about the motivation for this research.
10.31
For the named entity recognition task, we also find that KinyaBERT gives the best performance with the affix distribution regression variant performing best.
بالنسبة لمهمة named entity recognition، نجد أيضًا أن KinyaBERT يعطي أفضل أداء مع أداء متغير تراجع توزيع اللواحق بشكل أفضل.
Pour la tâche de reconnaissance d'entité nommée, nous constatons également que KinyaBERT fournit les meilleures performances avec la variante de régression de distribution d’affixes fonctionnant le mieux.
train_0024
speaker_001
10.81
For a morphological analyzer, we use finite state two level morphology principles with custom implementation that is tailored to the Kinyarwanda language.
2.85
A few remarks here for positional encoding.
وترد بعض الملاحظات هنا للتشفير الموضعي.
Quelques remarques ici pour l'encodage positionnel.
train_0025
speaker_001
9.92
The second task is Kinyarwanda named entity recognition benchmark, which is a high quality dataset that was annotated by trained native speakers.
7.08
These findings should motivate further research into morphology aware language pretrained language models.
يجب أن تحفز هذه النتائج المزيد من البحث في نماذج اللغة المدربة مسبقًا للغة التي تراعي المورفولوجيا.
Ces résultats devraient motiver davantage les recherches sur les modèles de langue préformée conscients de la morphologie.
train_0026
speaker_001
3.65
I'm a PhD student at the University of Massachusetts Amherst.
6.68
As you can see, Kinyarwanda words contains several morphemes that contain different information in them.
كما ترون، تحتوي كلمات اللغة الكينيرواندية على العديد من المورفيمات التي تحتوي على معلومات مختلفة فيها.
Comme vous pouvez le voir, les mots en kinyarwanda comprennent plusieurs morphèmes qui renferment différentes informations.
train_0027
speaker_001
4.14
The other option is to predict the affix probability vector.
6.19
For the GLUE benchmark, we find that KinyaBERT consistently outperforms baseline models.
بالنسبة لمعيار GLUE، نجد أن KinyaBERT يتفوق باستمرار على النماذج الأساسية.
Pour la référence GLUE, nous constatons que KinyaBERT dépasse systématiquement les modèles de référence.
train_0028
speaker_001
2.85
A few remarks here for positional encoding.
8.86
Also by looking at the low scores for finetuning, we find that KinyaBERT has better convergence in most cases.
أيضًا من خلال النظر في الدرجات المنخفضة للضبط الدقيق، نجد أن KinyaBERT لديه تقارب أفضل في معظم الحالات.
De plus, en examinant les faibles scores pour le raffinement, nous constatons que KinyaBERT a une meilleure convergence dans la plupart des cas.
train_0029
speaker_001
1.88
And embeddings for the stem.
4.75
Therefore, there is less gain from using pretrained language models.
لذلك، هناك استفادة أقل من استخدام نماذج لغة مدربة مسبقًا.
Par conséquent, il y a moins d'avantage à utiliser des modèles de langue préformée.
train_0030
speaker_001
9.52
So to conclude, this work has demonstrated the effectiveness of explicitly using morphological information in pretrained language models.
14.16
The proposed two tier transformer encoder architecture enables capturing morphological complexity morphological compositionality, which is an important aspect of morphologically rich languages.
تمكن بنية مشفر المحول المقترحة من مستويين من التقاط التعقيد المورفولوجي ونمذجة التركيبة المورفولوجية، وهو جانب مهم من اللغات الغنية مورفولوجيًا.
L’architecture à deux niveaux proposée permet de capturer la compositionnalité morphologique de la complexité morphologique, qui est un aspect important des langues morphologiquement riches.
train_0031
speaker_001
3.71
Here we show the average performance for ten finetuning runs.
10.31
For the named entity recognition task, we also find that KinyaBERT gives the best performance with the affix distribution regression variant performing best.
بالنسبة لمهمة named entity recognition، نجد أيضًا أن KinyaBERT يعطي أفضل أداء مع أداء متغير تراجع توزيع اللواحق بشكل أفضل.
Pour la tâche de reconnaissance d'entité nommée, nous constatons également que KinyaBERT fournit les meilleures performances avec la variante de régression de distribution d’affixes fonctionnant le mieux.
train_0032
speaker_001
1.68
My name is Antoine.
3.71
Here we show the average performance for ten finetuning runs.
ونعرض هنا متوسط الأداء لعشرة إجراءات للضبط الدقيق.
Ici, nous montrons la performance moyenne pour dix cycles de raffinement.
train_0033
speaker_001
4.97
This is because each of the morphemes occupies a known slot in the morphological model.
5.56
Essentially we have to predict both the stem and the affixes that are associated with the words.
وفي الأساس علينا أن نتنبأ بكل من الكلمة الجذعية واللواحق التي ترتبط بـ الكلمات.
Essentiellement, nous devons prévenir à la fois le radical et les affixes qui sont associés aux mots.
train_0034
speaker_001
15.04
This is because some morphological rules produce different surface forms that hide the exact lexical information, and BPE, which is solely based on the surface forms, does not have access to this lexical model.
9.52
So to conclude, this work has demonstrated the effectiveness of explicitly using morphological information in pretrained language models.
إذًا في الختام، أظهر هذا العمل فعالية الاستخدام الصريح للمعلومات المورفولوجية في نماذج اللغة المدربة مسبقًا.
Donc, pour conclure, ce travail a démontré l’efficacité de l’utilisation explicite des informations morphologiques dans les modèles de langue préformée.
train_0035
speaker_001
5.56
Essentially we have to predict both the stem and the affixes that are associated with the words.
12.8
All models are configured in the base architecture, which is about between a hundred and a hundred and ten million parameters, with Kinyarwanda with KinyaBERT using the least number of parameters.
يتم تكوين جميع النماذج في البنية الأساسية، والتي تتراوح بين مائة ومائة وعشرة ملايين من المعلمات، مع استخدام الكينيرواندية مع KinyaBERT باستخدام أقل عدد من المعلمات.
Tous les modèles sont configurés dans l’architecture de base, qui est d’environ cent à cent et dix millions de paramètres, avec le kinyarwanda avec KinyaBERT utilisant le plus petit nombre de paramètres.
train_0036
speaker_001
20.18
Due to the complex morphology that is expressed by most morphologically rich languages, the ubiquitous byte pair encoding tokenization algorithm that I used cannot extract the exact subword lexical units, meaning the morphemes, which are needed for effective representation.
10.63
The part of speech tagger uses a bidi bidirectional inference which improves upon the more often used Viterbi algorithm for decoding.
يستخدم جزء علامة الكلام استدلال ثنائي الاتجاه والذي يحسن من خوارزمية Viterbi الأكثر استخدامًا ل فك التشفير.
Le marqueur de partie de discours utilise une inférence bidirectionnelle bidi qui améliore le plus souvent l’algorithme Viterbi utilisé pour le décodage.
train_0037
speaker_001
20.18
Due to the complex morphology that is expressed by most morphologically rich languages, the ubiquitous byte pair encoding tokenization algorithm that I used cannot extract the exact subword lexical units, meaning the morphemes, which are needed for effective representation.
8.72
For this, we either group together affixes into a fixed number of sets and predict the set as a class label.
لهذا، نقوم إما بجمع اللواحق معًا في عدد ثابت من المجموعات ونتنبأ بالمجموعة كتصنيف فئة.
Pour cela, nous regroupons les affixes ensemble dans un nombre fixe d’ensembles et prévenons l’ensemble comme une étiquette de classe.
train_0038
speaker_001
5.03
I'll then talk about our experimental results, then finish with some conclusions.
7.43
But, all models had to cope with the same translation noise, and the relative performance between the models is still important to notice.
ولكن، كان على جميع النماذج التعامل مع نفس إزعاج الترجمة ولا يزال من المهم ملاحظة الأداء النسبي بين النماذج.
Mais tous les modèles ont dû faire face au même bruit de traduction, et la performance relative entre les modèles est toujours importante à remarquer.
train_0039
speaker_001
9.62
For example, here we have three Kinyarwanda words that have several morphemes in them, but the BPE algorithms cannot extract them.
11.12
We effectively model the morphology of all Kinyarwanda words, including verbals, nouns, demonstrative and possessive pronouns, numerals, and others.
نقوم بنمذجة مورفولوجيا جميع الكلمات الكينيرواندية، بما في ذلك الشفهيات، والأسماء، والضمائر التوضيحية وضمائر الملكية، والأرقام، وغيرها.
Nous modélisons efficacement la morphologie de tous les mots en kinyarwanda, y compris les verbes, les noms, les pronoms démonstratifs et possessifs, les chiffres et autres.
README.md exists but content is empty.
Downloads last month
15