=Paper=
{{Paper
|id=Vol-3159/T1-6
|storemode=property
|title=Leveraging Transformers for Hate Speech Detection in Conversational Code-Mixed Tweets
|pdfUrl=https://ceur-ws.org/Vol-3159/T1-6.pdf
|volume=Vol-3159
|authors=Zaki Mustafa Farooqi,Sreyan Ghosh,Rajiv Ratn Shah
|dblpUrl=https://dblp.org/rec/conf/fire/FarooqiGS21
}}
==Leveraging Transformers for Hate Speech Detection in Conversational Code-Mixed Tweets==
https://ceur-ws.org/Vol-3159/T1-6.pdf
Leveraging Transformers for Hate Speech Detection
in Conversational Code-Mixed Tweets
Zaki Mustafa Farooqi, Sreyan Ghosh and Rajiv Ratn Shah
Multimodal Digital Media Analysis Lab, Indraprastha Institute of Information Technology Delhi, India
Abstract
In the current era of the internet, where social media platforms are easily accessible for everyone, people
often have to deal with threats, identity attacks, hate, and bullying due to their association with a cast,
creed, gender, religion, or even acceptance or rejection of a notion. Existing works in hate speech
detection primarily focus on individual comment classification as a sequence labelling task and often
fail to consider the context of the conversation. The context of a conversation often plays a substantial
role when determining the author’s intent and sentiment behind the tweet. This paper describes the
system proposed by team MIDAS-IIITD for HASOC 2021 subtask 2, one of the first shared tasks focusing
on detecting hate speech from Hindi-English code-mixed conversations on Twitter. We approach this
problem using neural networks, leveraging the transformer’s cross-lingual embeddings and further fine-
tuning them for low-resource hate-speech classification in transliterated Hindi text. Our best performing
system, a hard voting ensemble of Indic-BERT, XLM-RoBERTa, and Multilingual BERT, achieved a macro
F1 score of 0.7253, placing us 1𝑠𝑡 on the overall leaderboard standings.
Keywords
Code-Mixed Languages, Hindi-English, Hate Speech, Transformers, Offensive Tweets
1. Introduction
In today’s world, hate speech is one of the major issues plaguing online social media websites.
Platforms like Twitter and Gab make it easier than ever before for a person to reach a large
audience quickly, which results in an increased temptation of users for inappropriate behavior
such as hate speech, causing potential damage to the social system and thus possessing major
threats which has already led to different types of crimes [1]. Human moderators manually
detecting hate speech online have been reported to go through trauma and mental issues. This
phenomenon necessitates automated hate speech detection as a crucial task.
A majority of the work on hate speech classification is constrained to the 𝐸𝑛𝑔𝑙𝑖𝑠ℎ language.
The inability of mono-lingual hate speech classifiers to detect the semantic cues in code-mixed
languages necessitates an efficient classifier that can detect offensive content automatically
from code-mixed languages. Hinglish (formed of the words spoken in Hindi language but
written in Roman script instead of the Devanagari script) extends its grammatical setup from
native Hindi, accompanied by many slurs, slang, and phonetic variations due to regional
Forum for Information Retrieval Evaluation, December 13-17, 2021, India
Envelope-Open zaki19048@iiitd.ac.in (Z. M. Farooqi); gsreyan@gmail.com (S. Ghosh); rajivratn@iiitd.ac.in (R. R. Shah)
GLOBE https://github.com/zmf0507 (Z. M. Farooqi); https://github.com/Sreyan88 (S. Ghosh); http://midas.iiitd.edu.in/
(R. R. Shah)
© 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
influence. Randomized spelling variations and multiple possible interpretations of Hinglish
words in different contextual situations make it extremely difficult to deal with for automated
classification. Another challenge worth considering in dealing with Hinglish is the demographic
divide between Hinglish users relative to total active users globally. This poses a severe limitation
as the tweet data in Hinglish language is a small fraction of the large pool of tweets generated,
necessitating the use of selective methods to process such tweets in an automated fashion.
Parent Tweet
भारतीय रे लवे के ऑ ीजन ए ेस अिभयान की 200वीं रे ल ने अपनी या ा पूरी
कर ली है । 10 अ रे लगािड़यां 784 मीिटक टन ऑ ीजन लेकर गितमान ह।दे श
भर म अभी तक 775 से अिधक टकरों के मा म से 12,630 मीिटक टन मेिडकल
ऑ ीजन प ं चाई गई है
Comment
@user Aaap sach me doctor ho ki aaise hi farji degree li
hai?
Reply
@user @user फज ही होगा
Figure 1: An example of conversation from the dataset where the parent tweet is not hateful but the
comment and reply are expressing implicit hate towards the user who posted parent tweet. The post
explains the amount of oxygen that has been supplied across the whole country. The hateful comment
says “Are you a genuine doctor or have you just acquired a fake degree?” while the hateful reply gives
an affirmative response by saying “it must be fake”. The green tick represents “NOT” comments while
the red cross represents “HOF” or hateful/offensive comments.
The context of the conversation plays a very crucial role in hate speech identification. We
acknowledge the fact that a comment categorized as hate speech may not always contain the
subject of hate on its own. As shown in figure 1, agreement or disagreement with a previous
comment or the overall ideology in the chain of the conversation might also induce hate towards
a particular target group. In addition to this, systems that can efficiently utilize the entire context
of the comment chain with a holistic understanding of the entire discussion may also help in
detecting “trigger comments” i.e., non-toxic comments in online discussions which lead to toxic
replies and implicitly help in mitigating bias in hate speech identification which remains a
long-standing problem in this domain [2, 3].
Hate Speech and Offensive Content Identification in English and Indo-Aryan Languages
(HASOC) 2021 [4] proposes two subtasks where subtask 1 aims towards identifying and dis-
criminating hate and profane tweets in English, Hindi, and Marathi, and subtask 2 aims towards
detecting hate tweets in conversations primarily in Hindi-English code-mixed texts. This paper
illustrates our key contribution to subtask 2 [5] of HASOC 2021 [4]. We present our system
based on ensemble of transformers to detect hate speech in code-mixed Hindi-English con-
versations, which helps us achieve the first place in the HASOC Subtask 2 final leaderboard
standings.
2. Related Work
Hate speech detection is a challenging task with literature including techniques such as
dictionary-based [6], distributional semantics [7] and recent literature exploring the power
of neural network architectures for the same [8]. However, a majority of the work done
on hate speech detection is constrained to the English language [9, 10, 11, 12, 13, 14, 15, 16]
with very limited work on other foreign languages [17, 18, 19, 20, 21] and code-switched text
[22, 23, 24, 25, 26]. Although Hinglish has been a major contributor to hate speech online, this
area has seen very little work with recent work exploring transformers [27] and author profiling
using graph neural networks [25]. Works like [22] uses a Convolutional Neural Network (CNN)
architecture togther with Glove embeddings and transfer learning, in one of the first attempts
to detect hate-speech online in the Hinglish language. Similar to our work, [28] explored XLM-
RoBERTa and achieved competitive results on the task of detecting hate-speech in Dravidian
languages. In HAOSC 2020 shared task, [29] also used XLM-RoBERTa to achieve the third rank
on the overall task of detecting offensive content in code-mixed Dravidian text. However, we
acknowledge the fact that XLM-RoBERTa was pre-trained on Hindi only text and our work
differs from theirs in which we also transliterate words in a different code to Hindi to solve the
problem of hate-speech detection.
Hate speech detection has branched into several sub-tasks like toxic span extraction [30, 31],
rationale identification [32] and hate target identification [20]. Though recent advancement
in the field of NLP has pushed the limits of hate speech identification, like transformers [25]
and graph neural networks [33, 25, 34] with people attempting to induce external knowledge
leveraging author profiling [25] or ideology [35] but using context of the conversation is still
a challenge with very little work exploring this problem. Context of the conversation plays a
huge role in hate speech identification with recent literature exploring both the structure and
effect of context [36, 37] for the same. One interesting and related direction of work described
in [38] relates to building systems that generate text which acts as hate speech interventions
in online discussion. Context can both help in detecting trigger comments [39] and implicitly
handle bias in hate speech identification which remains a long-standing problem in this domain
[2, 3]. To the best of our knowledge, all these works are constrained to the English language
with very little work on code-mixed Hindi-English and Hinglish text, considering the context
of the conversation.
3. Dataset
The dataset provided for this task has code-switched Hindi-English as well as Hinglish con-
versation chains taken from twitter as shown in Fig 1. This is a binary classification dataset
having two classes HOF and NOT. HOF denotes the tweet, comment, or reply which contains
hate, offensive, and profane content in itself or is supporting hate, whereas NOT denotes the
tweet, comment, or reply which does not contain any hate speech, profane, or offensive content.
More details about the dataset provided to us can be found in table 1. We have also provided
Avg. Comments which tells us about the average number of comments to a Parent Tweet in the
dataset since all Parent Tweets had at least one comment. We do not provide the average number
of replies for comments since all comments do not have replies.
Table 1
Original Dataset Statistics
Data Total Conversations HOF NOT Parent Tweets Total Comments Total Replies Avg. Comments
Train 5740 2841 2899 82 3778 1880 46
Test 1348 695 653 16 849 483 53
Table 2
Train-Validation-Test Distribution
Data Total Conversations HOF(Hateful/Offensive) NOT (neither Hateful nor Offensive)
Train 4592 2273 2319
Val 1148 568 580
Test 1348 695 653
Note : Val refers to Validation data .
For feeding data into our model, we flatten the given conversations into individual parent-
comment-reply unique conversation chains. As mentioned earlier, all parent tweets have atleast
one comment, but all comments do not have replies, so, a training instance might end with
a comment. For each instance, the final label assigned to the instance was the label of the
final comment in the conversation chain. Table 2 describes the dataset distribution with the
validation split used in our experiments in section 5.2 .
4. Methodology
Our methodology primarily involves fine-tuning the transformer models pre-trained on a
massive multilingual corpus. The following sections further describe the end-to-end approach
used in our experiments.
4.1. Data Pre-processing
We first start by concatenating the tweet and its comments and replies, if they are present, to
form the final text sequence. Our intuition behind this process is that this concatenation will
help the model understand the context better, especially in those cases where the comment or
reply may not be hateful but shows support for the hateful parent tweet. While concatenating
the tweets, we insert a new separator token “[SENSEP]” between the tweet , comment, and the
reply, to differentiate between one tweet and another. Post concatenation, we perform data
cleaning by removing hashtags, emojis, URLs and mentions from the tweets. However, we do
not remove punctuation and numbers to preserve the syntactic and semantic coherence of the
tweets.
Although the data is code-mixed, having both Hindi and English text, there were several
instances where Hindi text is present in Roman script. Since our models are pretrained on a
code-mixed corpus, it is essential to deal with Hindi text in the Roman script to make the whole
dataset consistent for training purposes. Therefore, we perform transliteration using AI4Bharat
library1 to convert the Hindi text in Roman script to Devanagari script.
Indic-BERT Logits Scores
Tweet Classifier
Roman - Devanagari
Hindi Transliteration, Logits Scores
XLM-RoBERTa Majority Voting
Comment Concat Removal of Softmax HOF/NOT
Classifier (Soft/Hard Voting)
Hashtags, Emojis,
Mentions
Multilingual BERT Logits Scores
Reply
Classifier
Figure 2: Overview of Ensemble Model. ’Concat’ refers to concatenation operation. ’Scores’ refer to
normalized logits/ probability scores .
4.2. Baseline Approach : Fine-tuning Transformer Models
We perform experiments with three different types of transformer models, which are described
below.
• Indic-BERT [40] is an ALBERT model pre-trained on a massive multilingual corpus
having 12 Indic languages such as Assamese, Bengali, English, Gujarati, Hindi, Kannada,
Malayalam, Marathi, Oriya, Punjabi, Tamil, Telugu [40] . The multilingual corpus has
about 9 billion tokens.
• Multilingual BERT (mBERT) is a BERT [41] model pretrained on Wikipedia data having
over 100 languages with a masked language modeling objective.
• XLM-Roberta (XLM-R) [42] is a transformer-based masked language model pretrained
on Common Crawl data having about 100 languages. It was proposed by Facebook and
happened to be one of the best-performing transformer models for multilingual tasks.
On top of these pre-trained transformer models, we add a dropout followed by a fully
connected layer of size two which takes in the transformer’s CLS token’s representation of size
768. The fully connected layer returns logits for the two classes, i.e., HOF and NOT, which are
then passed to a softmax layer to predict the class of the input text. This is explained as a part
of the ensemble model in figure 2 .
1
https://pypi.org/project/ai4bharat-transliteration/
4.3. Our Approach : Models Ensembling
We perform model ensembling on the top of our three fine-tuned models as explained in the
section 4.2. Since we had three different transformer models fine-tuned for our task, we decided
to combine the output of the three models using the majority voting system, i.e., Hard Voting
and Soft Voting. We denote our two ensemble models as Hard Voting Ensemble and Soft
Voting Ensemble. Hard Voting Ensemble model takes in the class predictions from each of
the fine-tuned transformer models and selects the class with maximum votes. Similarly, Soft
Voting Ensemble model takes in the class probabilities from each of the fine-tuned transformer
models and sum the same class probabilities and selects the class having higher probabilities
sum. Figure 2 shows the end-to-end model pipeline used in our experiment.
5. Experiments and Results
5.1. Experimental Setup
For fine-tuning our transformer models, we set a learning rate of 2e-5 for all our experiments
with AdamW [43] optimizer and a linear learning rate scheduler. We train our models with a
batch size of 8. Our experiments use the Hugging-Face Transformers library [44] for fine-tuning
all the pre-trained transformer models.
5.2. Evaluation with Validation data
In order to evaluate our models, we split the train dataset in an 80:20 ratio where 80% of the
data is the new train data, and the rest of 20% becomes the validation data as shown in table
2. The split is done in random order while maintaining the same class ratio in both train and
validation data. We train our models for ten epochs and keep track of the validation loss at each
epoch. For reporting results on validation data, We use the model checkpoint corresponding to
the epoch with minimum validation loss. At the same time, we make a note of that epoch for
which we later train our model on the whole train dataset in 5.3.
Table 3
Results obtained on Validation split
Model F1 Precision Recall Accuracy(%)
Indic-BERT 0.7150 0.7159 0.7154 71.51
Multilingual BERT 0.7438 0.7438 0.7439 74.39
XLM-RoBERTa 0.7262 0.7277 0.7268 72.64
Soft Voting Ensemble (Our approach) 0.7682 0.7687 0.7684 76.82
Hard Voting Ensemble (Our approach) 0.7621 0.7628 0.7624 76.21
Note : F1, Precision and Recall are Macro Scores
From Table 3 , we can see that our model ensembling approaches Soft Voting Ensemble
and Hard Voting Ensemble outperform rest of the baseline transformer models with Macro
F1 score of .7682 and .7621 respectively.
Table 4
Results obtained on Test Data
Submission Model F1 Precision Recall Accuracy(%)
NA Indic-BERT 0.6811 0.6881 0.6821 68.47
NA Multilingual BERT 0.7031 0.7031 0.7033 70.33
submit-1 XLM-RoBERTa 0.6970 0.6970 0.6970 69.73
submit-2 Soft Voting Ensemble (Our approach) 0.7223 0.7236 0.7222 72.32
submit-3 Hard Voting Ensemble (Our approach) 0.7253 0.7267 0.7251 72.62
Note : F1, Precision and Recall are Macro Scores
5.3. Evaluation with Test data
Following the results obtained in section 5.2 , we train our models on the whole train dataset for
the best number of epochs identified through the minimum validation loss in section 5.2 . These
epochs vary from 3 to 6 for each of the three transformer models discussed in 4.2. We submit the
test results for three of our models as marked in table 4 which reports the results obtained on the
test data. It can be inferred that the test results follow the same trend as was with the validation
data. However, Hard Voting Ensemble with Macro F1 of 0.7253 is slightly better than Soft
Voting Ensemble with Macro F1 score of 0.7223. However, the overall performance of the
model ensembling technique is better than the Indic-BERT, XLM-RoBERTa, and Multilingual
BERT by a significant margin where the highest possible Macro F1 score for baseline models is
0.7031.
6. Analysis
Macro F1 Score for Models
0.7
0.6
0.5
Macro F1 Score
0.4
0.3
0.2
0.1
0.0
T ERT Ta ble ble
-BER ual B oBER sem sem
Indic lt iling X LM-R t ing En t ing En
Mu Vo Vo
Soft Hard
Figure 3: Macro F1 Scores on Test Data
Figure 3 compares the results in table 4. It can be noted that the Indic-BERT model has
the lowest F1 score among all the baseline and ensemble models. Soft Voting Ensemble and
Hard Voting Ensemble models yield better results than all the baseline transformer models.
However, merely having a better F1 score is not enough since it is crucial to understand where
our approach fails and where it performs better than the baselines.
We start our error analysis with table 5 where we have shown the total number of misclassified
samples from each class. In addition to it, it has the percentage of total samples misclassified
from each class, which helps us develop a better understanding of the direction where models
are making more mistakes. Figure 4 shows the detailed results of the performance of each
model for samples of both classes. A closer look reveals that Multilingual BERT and XLM-
RoBERTa misclassify almost equal number of samples from both classes HOF and NOT where
the percentage ranges from 29.35% to 31.24%. However, the case is quite different for Indic-BERT,
where the misclassification rate is very high for the NOT class, which is 40.27% compared to
the misclassification rate of 23.30% for the HOF class. This could also be because Indic-BERT is
trained on 12 Indian languages and has a less diverse pre-training corpus than the other two
transformer models.
Table 5
Total Number and Percentage of Misclassified samples for HOF and NOT classes
Model Misclassified HOF % MR (HOF) Misclassified NOT % MR (NOT)
Indic-BERT 162 23.30 263 40.27
Multilingual BERT 207 29.78 193 29.55
XLM-RoBERTa 204 29.35 204 31.24
Soft Voting Ensemble 168 24.17 205 31.39
Hard Voting Ensemble 165 23.74 204 31.24
Note : % MR (HOF) and % MR (NOT) refer to percent misclassification rate for HOF and NOT classes
respectively. % MR (class) is the percentage of total misclassified samples of the class out of the total
samples of the class. These are calculated with total HOF samples count of 695 and total NOT samples
count of 653 in test data.
Indic-BERT Multilingual BERT XLM-RoBERTa Soft Voting Ensemble Hard Voting Ensemble
500 500 500
450 450
450 450 450
NOT 390 263 NOT 460 193 NOT 449 204 NOT 448 205 NOT 449 204
400 400
400 400 400
True Label
True Label
True Label
True Label
True Label
350 350 350 350 350
300 300 300 300
300
HOF 162 533 250 HOF 207 488 HOF 204 491 HOF 168 527 250 HOF 165 530 250
250
250
200 200 200
200
NOT HOF NOT HOF NOT HOF NOT HOF NOT HOF
Predicted Label Predicted Label Predicted Label Predicted Label Predicted Label
Figure 4: Confusion Matrix on Test Data for Baseline and Ensemble Models
As far as our ensemble models are concerned, we observe that the misclassification rate is
very balanced for both classes. To compare it with the baseline transformer models, we can see
that the ensemble models have the lowest misclassification rate for HOF class ranging between
23% to 24%, which was also the case with Indic-BERT, and similarly, the misclassification rate
for NOT class is also close to the lowest among the baseline models. This suggests that model
ensembling minimized the misclassification rate for both classes, and we can further conclude
that a single model’s mistake is likely to be corrected by the other two models in an ensemble
model. However, the ensemble models still make 7-8% more mistakes in identifying NOT class
compared to HOF class.
7. Conclusion
In this paper, we deal with a novel problem of detecting hateful tweets in twitter conversations
where the comment or reply might not be offensive and toxic but contributes to the hate
associated with the parent tweet. We performed a thorough experimental analysis with state-
of-the-art models such as XLM-RoBERTa, Indic-BERT, and Multilingual BERT to show that
pre-trained multi-lingual transformer models can achieve decent performance on this task. We
further demonstrate that this performance can be improved with model ensemble techniques
such as Soft Voting and Hard Voting. As this problem is dealt with for the first time, there could
be many ways to improve these numbers and build a more robust system by incorporating
other factors such as emojis and hashtags, which may equally or partially contribute to the
hatefulness of a tweet. Additionally, we aim to explore better architectures for taking into
account the context of a comment like parent comment and replies to judge the nature of the
comment. Author profiling also would be a potential area of research to detect implicit hate in
conversations.
References
[1] M. L. Williams, P. Burnap, A. Javed, H. Liu, S. Ozalp, Hate in the machine: anti-black and
anti-muslim social media posts as predictors of offline racially and religiously aggravated
crime, The British Journal of Criminology 60 (2020) 93–117.
[2] L. Dixon, J. Li, J. Sorensen, N. Thain, L. Vasserman, Measuring and mitigating unintended
bias in text classification, in: Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics,
and Society, 2018, pp. 67–73.
[3] D. Borkan, L. Dixon, J. Sorensen, N. Thain, L. Vasserman, Nuanced metrics for measuring
unintended bias with real data for text classification, in: Companion Proceedings of The
2019 World Wide Web Conference, 2019, pp. 491–500.
[4] S. Modha, T. Mandl, G. K. Shahi, H. Madhu, S. Satapara, T. Ranasinghe, M. Zampieri,
Overview of the HASOC Subtrack at FIRE 2021: Hate Speech and Offensive Content
Identification in English and Indo-Aryan Languages and Conversational Hate Speech, in:
FIRE 2021: Forum for Information Retrieval Evaluation, Virtual Event, 13th-17th December
2021, ACM, 2021.
[5] S. Satapara, S. Modha, T. Mandl, H. Madhu, P. Majumder, Overview of the HASOC Subtrack
at FIRE 2021: Conversational Hate Speech Detection in Code-mixed language , in: Working
Notes of FIRE 2021 - Forum for Information Retrieval Evaluation, CEUR, 2021.
[6] R. Guermazi, M. Hammami, A. B. Hamadou, Using a semi-automatic keyword dictionary
for improving violent web site filtering, in: 2007 Third International IEEE Conference on
Signal-Image Technologies and Internet-Based System, 2007, pp. 337–344. doi:1 0 . 1 1 0 9 /
SITIS.2007.137.
[7] N. Djuric, J. Zhou, R. Morris, M. Grbovic, V. Radosavljevic, N. Bhamidipati, Hate speech
detection with comment embeddings, in: Proceedings of the 24th International Conference
on World Wide Web, WWW ’15 Companion, Association for Computing Machinery, New
York, NY, USA, 2015, p. 29–30. URL: https://doi.org/10.1145/2740908.2742760. doi:1 0 . 1 1 4 5 /
2740908.2742760.
[8] P. Badjatiya, S. Gupta, M. Gupta, V. Varma, Deep learning for hate speech detection
in tweets, in: Proceedings of the 26th international conference on World Wide Web
companion, 2017, pp. 759–760.
[9] A.-M. Founta, C. Djouvas, D. Chatzakou, I. Leontiadis, J. Blackburn, G. Stringhini, A. Vakali,
M. Sirivianos, N. Kourtellis, Large scale crowdsourcing and characterization of twitter
abusive behavior, 2018. a r X i v : 1 8 0 2 . 0 0 3 9 3 .
[10] S. Carta, A. Corriga, R. Mulas, D. R. Recupero, R. Saia, A supervised multi-class multi-label
word embeddings approach for toxic comment classification., in: KDIR, 2019, pp. 105–112.
[11] H. H. Saeed, K. Shahzad, F. Kamiran, Overlapping toxic sentiment classification using deep
neural architectures, in: 2018 IEEE International Conference on Data Mining Workshops
(ICDMW), IEEE, 2018, pp. 1361–1366.
[12] A. Vaidya, F. Mai, Y. Ning, Empirical analysis of multi-task learning for reducing identity
bias in toxic comment detection, in: Proceedings of the International AAAI Conference
on Web and Social Media, volume 14, 2020, pp. 683–693.
[13] T. Tran, Y. Hu, C. Hu, K. Yen, F. Tan, K. Lee, S. Park, Habertor: An efficient and effective
deep hatespeech detector, 2020. a r X i v : 2 0 1 0 . 0 8 8 6 5 .
[14] H. Hitkul, R. R. Shah, P. Kumaraguru, S. Satoh, Maybe look closer? detecting trolling
prone images on instagram, in: 2019 IEEE Fifth International Conference on Multimedia
Big Data (BigMM), IEEE, 2019, pp. 448–456.
[15] Hitkul, K. Aggarwal, P. Bamdev, D. Mahata, R. R. Shah, P. Kumaraguru, Trawling for
trolling: A dataset, 2020. a r X i v : 2 0 0 8 . 0 0 5 2 5 .
[16] S. Ghosh, S. Lepcha, S. Sakshi, R. R. Shah, Speech toxicity analysis: A new spoken language
processing task, arXiv preprint arXiv:2110.07592 (2021).
[17] O. Kamal, A. Kumar, T. Vaidhya, Hostility detection in hindi leveraging pre-trained lan-
guage models, 2021. a r X i v : 2 1 0 1 . 0 5 4 9 4 .
[18] J. A. Leite, D. F. Silva, K. Bontcheva, C. Scarton, Toxic language detection in social
media for brazilian portuguese: New dataset and multilingual analysis, arXiv preprint
arXiv:2010.04543 (2020).
[19] A. Saroj, S. Pal, An indian language social media collection for hate and offensive speech, in:
Proceedings of the Workshop on Resources and Techniques for User and Author Profiling
in Abusive Language, 2020, pp. 2–8.
[20] V. Basile, C. Bosco, E. Fersini, D. Nozza, V. Patti, F. M. Rangel Pardo, P. Rosso, M. Sanguinetti,
SemEval-2019 task 5: Multilingual detection of hate speech against immigrants and women
in Twitter, in: Proceedings of the 13th International Workshop on Semantic Evaluation,
Association for Computational Linguistics, Minneapolis, Minnesota, USA, 2019, pp. 54–63.
URL: https://aclanthology.org/S19-2007. doi:1 0 . 1 8 6 5 3 / v 1 / S 1 9 - 2 0 0 7 .
[21] A. Ghosh Chowdhury, A. Didolkar, R. Sawhney, R. R. Shah, ARHNet - leveraging commu-
nity interaction for detection of religious hate speech in Arabic, in: Proceedings of the
57th Annual Meeting of the Association for Computational Linguistics: Student Research
Workshop, Association for Computational Linguistics, Florence, Italy, 2019, pp. 273–280.
URL: https://aclanthology.org/P19-2038. doi:1 0 . 1 8 6 5 3 / v 1 / P 1 9 - 2 0 3 8 .
[22] P. Mathur, R. Shah, R. Sawhney, D. Mahata, Detecting offensive tweets in hindi-english
code-switched language, in: Proceedings of the Sixth International Workshop on Natural
Language Processing for Social Media, 2018, pp. 18–26.
[23] R. Kapoor, Y. Kumar, K. Rajput, R. R. Shah, P. Kumaraguru, R. Zimmermann, Mind your
language: Abuse and offense detection for code-switched languages, in: Proceedings of
the AAAI Conference on Artificial Intelligence, volume 33, 2019, pp. 9951–9952.
[24] P. Mathur, R. Sawhney, M. Ayyar, R. Shah, Did you offend me? classification of offensive
tweets in hinglish language, in: Proceedings of the 2nd Workshop on Abusive Language
Online (ALW2), 2018, pp. 138–148.
[25] S. Chopra, R. Sawhney, P. Mathur, R. R. Shah, Hindi-english hate speech detection: Author
profiling, debiasing, and practical perspectives, in: Proceedings of the AAAI Conference
on Artificial Intelligence, volume 34, 2020, pp. 386–393.
[26] S. Kamble, A. Joshi, Hate speech detection from code-mixed hindi-english tweets using
deep learning models, 2018. a r X i v : 1 8 1 1 . 0 5 1 4 5 .
[27] T. Ranasinghe, S. Gupte, M. Zampieri, I. Nwogu, Wlv-rit at hasoc-dravidian-codemix-
fire2020: Offensive language identification in code-switched youtube comments, 2020.
arXiv:2011.00559.
[28] K. Yasaswini, K. Puranik, A. Hande, R. Priyadharshini, S. Thavareesan, B. R. Chakravarthi,
IIITT@DravidianLangTech-EACL2021: Transfer learning for offensive language detection
in Dravidian languages, in: Proceedings of the First Workshop on Speech and Language
Technologies for Dravidian Languages, Association for Computational Linguistics, Kyiv,
2021, pp. 187–194. URL: https://aclanthology.org/2021.dravidianlangtech-1.25.
[29] A. Baruah, K. A. Das, F. A. Barbhuiya, K. Dey, Iiitg-adbu@hasoc-dravidian-codemix-
fire2020: Offensive content detection in code-mixed dravidian text, 2021. a r X i v : 2 1 0 7 . 1 4 3 3 6 .
[30] J. Pavlopoulos, L. Laugier, J. Sorensen, I. Androutsopoulos, Semeval-2021 task 5: Toxic
spans detection, in: Proceedings of the 15th International Workshop on Semantic Evalua-
tion, 2021.
[31] S. Ghosh, S. Kumar, Cisco at semeval-2021 task 5: What’s toxic?: Leveraging transformers
for multiple toxic span extraction from online comments, 2021. a r X i v : 2 1 0 5 . 1 3 9 5 9 .
[32] B. Mathew, P. Saha, S. M. Yimam, C. Biemann, P. Goyal, A. Mukherjee, Hatexplain: A
benchmark dataset for explainable hate speech detection, 2020. a r X i v : 2 0 1 2 . 1 0 2 8 9 .
[33] P. Mishra, M. D. Tredici, H. Yannakoudakis, E. Shutova, Abusive language detection with
graph convolutional networks, 2019. a r X i v : 1 9 0 4 . 0 4 0 7 3 .
[34] M. Das, P. Saha, R. Dutt, P. Goyal, A. Mukherjee, B. Mathew, You too brutus! trapping
hateful users in social media: Challenges, solutions insights, 2021. a r X i v : 2 1 0 8 . 0 0 5 2 4 .
[35] J. Qian, M. ElSherief, E. Belding, W. Y. Wang, Hierarchical cvae for fine-grained hate speech
classification, 2018. a r X i v : 1 8 0 9 . 0 0 0 8 8 .
[36] J. Pavlopoulos, J. Sorensen, L. Dixon, N. Thain, I. Androutsopoulos, Toxicity detection:
Does context really matter?, 2020. a r X i v : 2 0 0 6 . 0 0 9 9 8 .
[37] M. Saveski, B. Roy, D. Roy, The structure of toxic conversations on twitter, 2021.
arXiv:2105.11596.
[38] J. Qian, A. Bethke, Y. Liu, E. Belding, W. Y. Wang, A benchmark dataset for learning to
intervene in online hate speech, arXiv preprint arXiv:1909.04251 (2019).
[39] H. Almerekhi, H. Kwak, J. Salminen, B. J. Jansen, Are these comments triggering? pre-
dicting triggers of toxicity in online discussions, WWW ’20, Association for Comput-
ing Machinery, New York, NY, USA, 2020. URL: https://doi.org/10.1145/3366423.3380074.
doi:1 0 . 1 1 4 5 / 3 3 6 6 4 2 3 . 3 3 8 0 0 7 4 .
[40] D. Kakwani, A. Kunchukuttan, S. Golla, G. N.C., A. Bhattacharyya, M. M. Khapra, P. Ku-
mar, IndicNLPSuite: Monolingual Corpora, Evaluation Benchmarks and Pre-trained
Multilingual Language Models for Indian Languages, in: Findings of EMNLP, 2020.
[41] J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep bidirectional
transformers for language understanding, 2019. a r X i v : 1 8 1 0 . 0 4 8 0 5 .
[42] A. Conneau, K. Khandelwal, N. Goyal, V. Chaudhary, G. Wenzek, F. Guzmán, E. Grave,
M. Ott, L. Zettlemoyer, V. Stoyanov, Unsupervised cross-lingual representation learning at
scale, 2020. a r X i v : 1 9 1 1 . 0 2 1 1 6 .
[43] I. Loshchilov, F. Hutter, Decoupled weight decay regularization, 2019. a r X i v : 1 7 1 1 . 0 5 1 0 1 .
[44] T. Wolf, L. Debut, V. Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf,
M. Funtowicz, J. Davison, S. Shleifer, P. von Platen, C. Ma, Y. Jernite, J. Plu, C. Xu, T. L. Scao,
S. Gugger, M. Drame, Q. Lhoest, A. M. Rush, Transformers: State-of-the-art natural lan-
guage processing, in: Proceedings of the 2020 Conference on Empirical Methods in Natural
Language Processing: System Demonstrations, Association for Computational Linguistics,
Online, 2020, pp. 38–45. URL: https://www.aclweb.org/anthology/2020.emnlp-demos.6.