txt/2106.14625.txt

AMU-EURANOV A at CASE 2021 Task 1: Assessing the stability of
multilingual BERT
L´eo Bouscarrat1;2, Antoine Bonnefoy1, C´ecile Capponi2, Carlos Ramisch2
1EURA NOV A, Marseille, France
2Aix Marseille Univ, Universit ´e de Toulon, CNRS, LIS, Marseille, France
fleo.bouscarrat, antoine.bonnefoy [email protected]
fleo.bouscarrat, cecile.capponi, carlos.ramisch [email protected]
Abstract
This paper explains our participation in task 1
of the CASE 2021 shared task. This task is
about multilingual event extraction from news.
We focused on sub-task 4, event information
extraction. This sub-task has a small training
dataset and we fine-tuned a multilingual BERT
to solve this sub-task. We studied the instabil-
ity problem on the dataset and tried to mitigate
it.
1 Introduction
Event extraction is becoming more and more impor-
tant as the number of online news increases. This
task consists of extracting events from documents,
especially news. An event is defined by a group of
entities that give some information about the event.
Therefore, the goal of this task is to extract, for
each event, a group of entities that define the event,
such as the place and time of the event.
This task is related but still different from named
entity recognition (NER) as the issue is to group
the entities that are related to the same event, and
differentiate those related to different events. This
difference makes the task harder and also compli-
cates the annotation.
In the case of this shared task, the type of events
to extract is protests (H ¨urriyeto ˘glu et al., 2021a,b).
This shared task is in the continuation of two previ-
ous shared tasks at CLEF 2019 (H ¨urriyeto ˘glu et al.,
2019) and AESPEN (H ¨urriyeto ˘glu et al., 2020).
The first one deals with English event extraction
with three sub-tasks: document classification, sen-
tence classification, and event information extrac-
tion. The second focuses on event sentence co-
reference identification, whose goal is to group
sentences related to the same events.
This year, task 1 is composed of the four afore-
mentioned tasks and adds another difficulty: multi-
linguality. This year’s data is available in English,Spanish, and Portuguese. Thus, it is important to
note that there is much more data in English than
in the other languages. For the document classi-
fication sub-task, to test multilingual capabilities,
Hindi is available on the testing set only.
We have mainly focused on the last sub-task
(event information extraction), but we have also
submitted results for the first and second sub-tasks
(document and sentence classification). We used
multilingual BERT (Devlin et al., 2019), hence-
forth M-BERT, which is a model known to obtain
near state-of-the-art results on many tasks. It is also
supposed to work well for zero-or-few-shot learn-
ing on different languages (Pires et al., 2019). We
will see the results on these sub-tasks, especially
for sub-task 4 where the training set available for
Spanish and Portuguese is small.
Thus, one of the issues with transformer-based
models such as M-BERT is the instability on small
datasets (Dodge et al., 2020; Ruder, 2021). The
instability issue is the fact that by changing some
random seeds before the learning phase but using
the same architecture, data and hyper-parameters
the results can have a great variance. We will look
at some solutions to mitigate this issue, and how
this issue is impacting our results for sub-task 4.1
2 Tasks and data
Sub-tasks 1 and 2 can be seen as binary sequence
classification, where the goal is to say if a given
sequence is part of a specific class. In our case, a
classifier must predict whether a document contains
information about an event for sub-task 1 or if a
sentence contains information about an event for
sub-task 2.
Document and sentence classification tasks, sub-
tasks 1 and 2, are not our main research interest.
1Our code is available here: https://github.com/
euranova/AMU-EURANOVA-CASE-2021arXiv:2106.14625v2  [cs.CL]  4 Aug 2021Figure 1: Example of a snippet from sub-task 4.
Moreover, the datasets provided for these tasks
are less interesting (reasonable amount of training
data).
On the other hand, sub-task 4 not only has less
training data available but also requires more fine-
grained token-based prediction. The goal of sub-
task 4 is to extract event information from snippets
that contain sentences speaking about the same
event. H ¨urriyeto ˘glu et al. (2019) have defined that
an event has the following information classes (ex-
ample in Figure 1):
•Time, which indicates when the protest took
place,
•Facility name, which indicates in which facil-
ity the protest took place,
•Organizer, which indicates who organized the
protest,
• Participant, which indicates who participated
in the protest,
•Place, which indicates where the protest took
place in a more general area than the facility
(city, region, ...),
•Target, which indicates against whom or what
the protest took place,
•Trigger, which is a specific word or group of
words that indicate that a protest took place
(examples: protested, attack, ...),
Thus, not all the snippets contain all the classes,
and they can contain several times the same classes.
Each information can be composed of one or sev-
eral adjacent words. Each snippet contains infor-
mation related to one and only one event.
As the data is already separated into groups
of sentences related to the same event, our ap-
proach consists of considering a task of named
entity recognition with the aforementioned classes.
Multilingual BERT has already been used for multi-
lingual named entity recognition and showed great
results compared to state-of-the-art models (Hakala
and Pyysalo, 2019).The data is in BIO format (Ramshaw and Mar-
cus, 1995), where each word has a B tag or an I tag
of a specific class or an O tag. The B tag means
beginning and marks the beginning of a new entity.
The tag I means inside, which has to be preceded
by another I tag or a B tag, and marks that the word
is inside an entity but not the first word of the entity.
Finally, the O-tag means outside, which means the
word is not part of an entity.
3 System overview
Our model is based on pre-trained multilingual
BERT (Devlin et al., 2019). This model has been
pretrained on multilingual Wikipedia texts. To bal-
ance the fact that the data is not equally distributed
between all the languages the authors used expo-
nential smoothed weighting to under-sample the
most present languages and over-sample the rarest
ones. This does not perfectly balance all the lan-
guages but it reduces the impact of low-resourced
languages.
The authors of the M-BERT paper shared the
weights of a pretrained model that we use to do fine-
tuning. Fine-tuning a model consists of taking an
already trained model on a specific task and using
this model as a starting point of the training for the
task of interest. This approach has reached state-
of-the-arts in numerous tasks. In the case of M-
BERT, the pre-training tasks are Masked Language
Modeling (MLM) and Next Sentence Prediction
(NSP).
To be able to learn our task, we add a dense layer
on top of the outputs of M-BERT and learn it during
the fine-tuning. All our models are fine-tuning all
the layers of M-BERT.
The implementation is the one from Hugging-
Face’s ‘transformers’ library (Wolf et al., 2020). To
train it on our data, the model is fine-tuned on each
sub-task.
3.1 Sub task 1 and 2
For sub-tasks 1 and 2, we approach these tasks
as binary sequence classification, as the goal is to
predict whether or not a document (sub-task 1) orsentence (sub-task 2) contains relevant information
about a protest event. Thus the size of the output
of the dense layer is 2. We then perform an argmax
on these values to predict a class. We use the base
parameters in HuggingFace’s ’transformers’ library.
The loss is a cross-entropy, the learning rate is
handled by an AdamW optimizer (Loshchilov and
Hutter, 2019) and the activation function is a gelu
(Hendrycks and Gimpel, 2016). We use a dropout
of 10% for the fully connected layers inside M-
BERT and the attention probabilities.
One of the issues with M-BERT is the limited
length of the input, as it can only take 512 tokens,
which are tokenized words. M-BERT uses the
wordpiece tokenizer (Wu et al., 2016). A token is
either a word if the tokenizer knows it, if it does not
it will separate it into several sub-tokens which are
known. For sub-task 1, as we are working with en-
tire documents, it can be frequent that a document
is longer than this limit and has to be broken down
into several sub-documents. To retain contexts in
each sub-documents we use an overlap of 150 to-
kens, which means between two sub-documents,
they will have 150 tokens in common. Our method
to output a class, in this case, is as follows:
• tokenize a document,
•if the tokenized document is longer than
the 512-tokens limit, create different sub-
documents with 150-tokens overlaps between
each sub-document,
• generate a prediction for each sub-document,
•average all the predictions from sub-
documents originated from the same docu-
ment,
• take the argmax of the final prediction.
3.2 Sub-task 4
For sub-task 4, our approach is based on word
classification where we predict a class for each
word of the documents.
One issue is that as words are tokenized and can
be transformed into several sub-tokens we have to
choose how to choose the prediction of a multi-
token word. Our approach is to take the prediction
of the first token composing a word as in Hakala
and Pyysalo (2019).
We also have to deal with the input size as some
documents are longer than the limit. In this case,we separate them into sub-documents with an over-
lap of 150. Our approach is:
• tokenize a document,
•if the tokenized document is longer than
the 512-tokens limit, create different sub-
documents with 150-tokens overlaps between
each sub-document,
• generate a prediction for each sub-document,
•reconstruct the entire document: take the first
and second sub-documents, average the pre-
diction for the same tokens (from the overlap),
keep the prediction for the others, then use
the same process with the obtained document
and the next sub-document. As the size of
each sequence is 512 and the overlap is only
150, no tokens can be in more than 2 different
sequences,
•take the argmax of the final prediction for each
word.
3.2.1 Soft macro-F1 loss
We used a soft macro-F1 loss (Lipton et al., 2014).
This loss is closer than categorical cross-entropy on
BIO labels to the metric used to evaluate systems
in the shared task. The main issue with F1 is its
non-differentiability, so it cannot be used as is but
must be modified to become differentiable. The F1
score is based on precision and recall, which in turn
are functions of the number of true positives, false
positives, and false negatives. These quantities are
usually defined as follows:
tp=X
i2tokens(pred (i)true (i))
fp=X
i2tokens(pred (i)(1true (i)))
fn=X
i2tokens((1pred (i))true (i))
With:
•tokens , the list of tokens in a document,
•true(i) , 0 if the true label of the token i is of
the negative class, 1 if the true label is of the
positive class
•pred(i) , 0 if the predicted label of the token i
is of the negative class, 1 if the predicted label
is of the positive classAs we use macro-F1 loss, we compute the F1
score for each class where the positive class is the
current class and negative any other class, e.g. if
the reference class is B-trigger, then true(i)=1 for
B-trigger and true(i)=0 for all other classes when
macro-averaging the F1.
We replace the binary function pred(i) by a func-
tion outputting the predicted probability of the to-
ken i to be of the positive class:
soft tp=X
i2tokens(proba (i)true (i))
soft fp=X
i2tokens(proba (i)(1true (i)))
soft fn=X
i2tokens((1proba (i))true (i))
With proba(i) outputting the probability of the
token i to be of the positive class, this probability is
the predicted probability resulting from the softmax
activation of the fine-tuning network.
Then we compute, in a similar fashion as a nor-
mal F1, the precision and recall using the soft defi-
nitions of the true positive, false positive, and false
negative. And finally we compute the F1 score with
the given precision and recall. As a loss function is
a criterion to be minimized whereas F1 is a score
that we would like to maximize, the final loss is
1F1.
3.2.2 Recommendation for improved stability
A known problem of Transformers-based models
is the training instability, especially with small
datasets (Dodge et al., 2020; Ruder, 2021). Dodge
et al. (2020) explain that two elements that have
much influence on the stability are the data order
and the initialization of the prediction layer, both
controlled by pseudo-random numbers generated
from a seed. To study the impact of these two el-
ements on the models’ stability, we freeze all the
randomness on the other parts of the models and
change only two different random seeds:
•the data order, i.e. the different batches and
their order. Between two runs the model will
see the same data during each epoch but the
batches will be different, as the batches are
built beforehand and do not change between
epochs,
•the initialization of the linear layer used to
predict the output of the model.Another recommendation to work with
Transformers-based models and small data made
by Mosbach et al. (2021) is to use smaller learning
rates but compensating with more epochs. We have
taken this into account during the hyper-parameter
search.
Ruder (2021) recommend using behavioral fine-
tuning to reduce fine-tuning instabilities. It is sup-
posed to be especially helpful to have a better ini-
tialization of the final prediction layer. It has also al-
ready been used on named entity recognition tasks
(Broscheit, 2019) and has shown that it has im-
proved results for a task with a very small training
dataset. Thus, to do so, we need a task with the
same number of classes, but much larger training
datasets. As we did not find such a task, we de-
cided to fine-tune our model on at least the different
languages we are working with, English, Spanish
and Portuguese. We used named entity recognition
datasets and kept only three classes in common in
all the datasets: person, organization, and location.
These three types of entities can be found in the
shared task.
To perform this test, the training has been done
like that:
•the first fine-tuning is done on the concatena-
tion of NER datasets in different languages,
once the training is finished we save all the
weights of the model,
•we load the weights of the previous model,
except for the weights of the final prediction
layer which are randomized with a given seed,
•we train the model on the dataset of the shared
task.
4 Experimental setup
4.1 Data
The dataset of the shared task is based on articles
from different newspapers in different languages.
More information about this dataset can be found
in (H ¨urriyeto ˘glu et al., 2021a)
For the final submissions of sub-tasks 1, 2, and 4
we divided the dataset given for training purposes
into two parts with 80% for training and 20% for
evaluation during the system training phase. We
then predicted the data given for testing purposes
during the shared task evaluation phase. The quan-
tity of data for each sub-task and language can be
found in Table 1. We can note that the majority ofSub-task English Spanish Portuguese
Sub-task 1 9,324 1,000 1,487
Sub-task 2 22,825 2,741 1,182
Sub-task 4 808 33 30
Table 1: Number of elements for each sub-task for each
language in the data given for training purposes. Docu-
ments for sub-task 1, sentences for sub-task 2, snippet
(group of sentences about one event) for sub-task 4.
Dataset Train Eval Test
CoNLL 2003 14,041 3,250 3,453
CoNLL 2002 8,324 1,916 1,518
HAREM 121 8 128
Table 2: Number of elements for each dataset used in
the behavioral fine-tuning in each split.
the data is in English. Spanish and Portuguese are
only a small part of the dataset.
For all the experiments made on sub-task 4, we
divided the dataset given for training purposes into
three parts with 60% for training, 20% for evaluat-
ing and 20% for testing.
To be able to do our approach of behavioral fine-
tuning, we needed some Named Entity Recognition
datasets in English, Spanish and Portuguese. For
English we used the CoNLL 2003 dataset (Tjong
Kim Sang and De Meulder, 2003), for Spanish the
Spanish part of the CoNLL 2002 dataset (Tjong
Kim Sang, 2002) and for Portuguese the HAREM
dataset (Santos et al., 2006). Each of these datasets
had already three different splits for training, devel-
opment and test. Information about their size can
be found in Table 2.
The dataset for Portuguese is pretty small com-
pared to the two others, but the impact of the size
can be interesting to study.
4.2 Hyper-parameter search
For sub-task 4, we did a hyper-parameter search
to optimize the results. We used Ray Tune (Liaw
et al., 2018) and the HyperOpt algorithm Bergstra
et al. (2013). We launched 30 different trainings,
all the information about the search space and the
hyper-parameters can be found in A.1. The goal is
to optimize the macro-F1 on the evaluation set.
Our goal was to find a set of hyper-parameters
that performs well to use always the same in the
following experiments. We also wanted to evaluate
the impacts of the hyper-parameters on the training.4.3 Behavioral fine-tuning
For the first part of the behavioral fine-tuning,
we trained an M-BERT model on the three NER
datasets for one epoch. We only learn for one epoch
for timing issues, as the learning on this datasets
takes several hours. We then fine-tune the resulting
models with the best set of hyper-parameters found
with the hyper-parameter search.
4.4 Stability
To study the stability of the model and the impact
of behavioral fine-tuning we made 6 sets of experi-
ments with 20 experiments in each set:
•normal fine-tuning with random data order
and frozen initialization of final layer,
•normal fine-tuning with frozen data order and
random initialization of final layer,
•normal fine-tuning with random data order
and random initialization of final layer,
•behavioral fine-tuning with random data order
and frozen initialization of final layer,
•behavioral fine-tuning with frozen data order
and random initialization of final layer,
•behavioral fine-tuning with random data order
and random initialization of final layer,
Once again it is important to note that what we
called behavioral fine-tuning is different from be-
havioral fine-tuning as proposed by Ruder (2021),
as we reset the final layer. Only the weights of all
the layers of M-BERT are modified.
For each set of experiments we will look at
the average of the macro-F1, as implemented in
Nakayama (2018), and the standard deviation of
the macro-F1 on the training dataset, on the evalua-
tion dataset, and on three different test datasets, one
for each language. Thus we will be able to assess
the importance of the instability, if our approach to
behavioral fine-tuning helps to mitigate it and if it
has similar results across the languages.
We can also note that in our implementation
the batches are not randomized. They are built
once before the learning phase and do not change,
neither in content nor order of passage, between
each epoch.Figure 2: (Top) Parallel coordinates plot of the 30 experiments on sub-task 4 during the hyper-parameter search in
function of the value of the hyper-parameters and the value of the F1 on the evaluation set. Each line represents an
experiment, and each column a specific hyper-parameter, except the last which is the value of the metric. (Bottom)
Same plot with the worst results removed to have a better view of the best results.
5 Results
5.1 Hyper-parameter search
The results of the hyper-parameter search can be
seen in Figure 2. On the top pictures which repre-
sent the 30 experiments, we can see that a specific
hyper-parameter seems to impact the worst results
(in blue). This parameter is the learning rate, we
can see it in the red box on the top image, all the
blue lines are at the bottom, which means these
experiments had a small learning rate. It seems
that we obtain the best results with a learning rate
around 5e-05 (0.00005), lower than 1e-06 seems to
give bad results.
We can then focus on the bottom picture, with
the same type of plot but with the worst results
removed. Another hyper-parameter that seems to
have an impact is the number of training epochs,
40 seems better than 20. We use a high number of
epochs as recommended by Mosbach et al. (2021)
to limit the instability. Beyond the learning rate and
number of epochs, it is then hard to find impactful
hyper-parameters.
Finally, the set of hyper-parameters that has been
selected is:• Adafactor: True
• Number of training epochs: 40
• Adam beta 2: 0.99
• Adam beta 1: 0.74
• Maximum gradient norm: 0.17
• Adam epsilon: 3e-08
• Learning rate: 5e-05
• Weight decay: 0.36
For the stability experiments, the number of
training epochs have been reduced to 20 for speed
purposes. For the first part of the behavioral fine-
tuning, the learning rate has been set to 1e-05 as
more data were available.
5.2 Behavioral fine-tuning
The results on the test dataset of each model after
one epoch of training can be found in Table 5.
We could not compare to state-of-the-art NER
models on these three datasets as we do not take all
the classes (classes such as MISC were removedData Init layer Train Eval Test EN Test ES Test PT
NRand Fix 86.11 (1.08) 69.34 (1.01) 71.80 (.85) 54.33 (3.43) 73.14 ( 1.96)
Fix Rand 86.88 (.53) 70.03 (.63) 71.68 ( .53)55.02 (3.28) 74.51 (2.41)
Rand Rand 86.63 (1.08) 69.56 (.97) 71.94 (.72) 54.73 (3.44) 74.08 (3.37)
BRand Fix 85.79 (.97) 69.32 (1.00) 71.60 (.54) 54.69 (2.99) 74.01 (2.92)
Fix Rand 86.20 (.55) 69.57 ( .51)71.80 (.58) 53.97 (3.90) 74.50 (2.67)
Rand Rand 86.11 (.87) 69.40 (.80) 71.85 (.73) 55.51 (2.82)74.97 (2.66)
Table 3: Average macro-F1 score, higher is better (standard deviation, lower is better) of the 20 experiments with
the specified setup. N means normal fine-tuning and B behavioral fine-tuning. Data means data order and Init layer
means initialization of the final layer. Rand means random, and fix refers to frozen.
English Spanish Portuguese Hindi
Sub-task 1 53.46 (84.55) 46.47 (77.27) 46.47 (84.00) 29.66 (78.77)
Sub-task 2 75.64 (85.32) 76.39 (88.61) 81.61 (88.47) /
Sub-task 4 69.96 (78.11) 56.64 (66.20) 61.87 (73.24) /
Table 4: Score of our final submissions for each sub-task, in parenthesis the score achieved by the best scoring
team on each sub-task.
Dataset Test macro-F1
CoNLL 2003 89.8
CoNLL 2002 86.1
HAREM 76.1
Table 5: Macro-F1 score of the NER task on the test
split of each dataset used in behavioral fine-tuning after
training the base M-BERT for 1 epoch.
before the learning phase). The metrics used on
these datasets are not by classes, so the comparison
cannot be made. However, the results are already
much better than what a random classifier would
output, thus the weights of the models should al-
ready be better than the weights of the base model.
5.3 Stability
The results of the different sets of experiments can
be found in Table 3. First, we can see that the dif-
ference between behavioral fine-tuning and normal
fine-tuning is not important enough to say one is
better than the other. We can also note that the
standard deviation is small for English, but not
negligible for Spanish and Portuguese.
5.4 Final submission
The results of the final submissions can be found
in Table 4. We can see that our results are lower
than the best results, especially for sub-task 1 with
a difference of between 30 to 50 macro-F1 score
depending on the language, whereas for sub-tasks2 and 4 the difference is close to 10 macro-F1 score
for all the languages.
6 Conclusion
6.1 Sub-task 1 and 2
As we can see in Table 4, our final results for sub-
task 1 are much lower than the best results, but for
sub-task 2 the difference is smaller. This is interest-
ing as the tasks are pretty similar, thus expected the
difference between our results and the best results
to be of the same magnitude.
One explanation could be our approach to han-
dle documents longer than the input of M-BERT.
We have chosen to take the average of the sub-
documents, but if one part of a document contains
an event the entire document does too. We may
have better results looking if one sub-document at
least is considered as having an event.
It is then hard to compare to other models as we
have chosen to use one model for all the languages
and we do not know the other approaches.
6.2 Sub-task 4
For sub-task 4 we have interesting results for all
the languages, even for Spanish and Portuguese,
as we were not sure that we could learn this task
in a supervised fashion with the amount of data
available. In a further study, we could compare our
results with results obtained by fine-tuning mono-
lingual models, where we fine-tune one model for
each language with only the data of one language.This could show the impact of having data if using
a multilingual model instead of several monolin-
gual models improves or not the results. We do not
expect good results for Spanish and Portuguese as
the training dataset is pretty limited. The results
seem to comfort the claim of (Pires et al., 2019)
that M-BERT works well for few-shot learning on
other languages.
The other question for sub-task 4 was about in-
stability. In Table 3 we can see that the instability is
way more pronounced for Spanish and Portuguese.
It seems logical as we have fewer data available
in Spanish and Portuguese than in English. The
standard deviation for Spanish and Portuguese is
large and can have a real impact on the final re-
sults. Finding good seeds could help to improve
the results for Spanish and Portuguese.
Furthermore, our approach of behavioral fine-
tuning did not help to reduce the instabilities. It
was expected that one of the sources of the insta-
bility is the initialization of the prediction, and in
our approach, the initialization of this layer is still
random. In our approach, we only fine-tune the
weights of M-BERT. This does not seem to work
and reinforces the advice of Ruder (2021) that us-
ing behavioral fine-tuning is more useful for having
a good initialization of the final prediction layer.
On the two sources of randomness we studied,
data order seems the most impactful for English,
where we have more data. Nonetheless, for Span-
ish and Portuguese, the two sources have a large
impact. In a further study, we could see how the
quantity of data helps to decrease the impact of
these sources of instabilities.
For the final submissions, the macro-F1 score
for English and Portuguese is beneath the average
macro-F1 score we found during our development
phases. This could be due to bad seeds for random-
ness or because the splits are different. We did not
try to find the best-performing seeds for the final
submissions.
Acknowledgments
We thank Damien Fourrure, Arnaud Jacques, Guil-
laume Stempfel and our anonymous reviewers for
their helpful comments.
References
James Bergstra, Daniel Yamins, and David Cox. 2013.
Making a science of model search: Hyperparameteroptimization in hundreds of dimensions for vision ar-
chitectures. In International conference on machine
learning , pages 115–123. PMLR.
Samuel Broscheit. 2019. Investigating entity knowl-
edge in BERT with simple neural end-to-end en-
tity linking. In Proceedings of the 23rd Confer-
ence on Computational Natural Language Learning
(CoNLL) , pages 677–685, Hong Kong, China. Asso-
ciation for Computational Linguistics.
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and
Kristina Toutanova. 2019. Bert: Pre-training of
deep bidirectional transformers for language under-
standing. In Proceedings of the 2019 Conference of
the North American Chapter of the Association for
Computational Linguistics: Human Language Tech-
nologies, Volume 1 (Long and Short Papers) , pages
4171–4186.
Jesse Dodge, Gabriel Ilharco, Roy Schwartz, Ali
Farhadi, Hannaneh Hajishirzi, and Noah Smith.
2020. Fine-tuning pretrained language models:
Weight initializations, data orders, and early stop-
ping. arXiv preprint arXiv:2002.06305 .
Kai Hakala and Sampo Pyysalo. 2019. Biomedical
named entity recognition with multilingual BERT.
InProceedings of The 5th Workshop on BioNLP
Open Shared Tasks , pages 56–61, Hong Kong,
China. Association for Computational Linguistics.
Dan Hendrycks and Kevin Gimpel. 2016. Gaus-
sian error linear units (gelus). arXiv preprint
arXiv:1606.08415 .
Ali H ¨urriyeto ˘glu, Osman Mutlu, Farhana Ferdousi
Liza, Erdem Y ¨or¨uk, Ritesh Kumar, and Shyam
Ratan. 2021a. Multilingual protest news detection
- shared task 1, case 2021. In Proceedings of the 4th
Workshop on Challenges and Applications of Auto-
mated Extraction of Socio-political Events from Text
(CASE 2021) , online. Association for Computational
Linguistics (ACL).
Ali H ¨urriyeto ˘glu, Hristo Tanev, Vanni Zavarella, Jakub
Piskorski, Reyyan Yeniterzi, and Erdem Y ¨or¨uk.
2021b. Challenges and applications of automated
extraction of socio-political events from text (case
2021): Workshop and shared task report. In
Proceedings of the 4th Workshop on Challenges
and Applications of Automated Extraction of Socio-
political Events from Text (CASE 2021) , online. As-
sociation for Computational Linguistics (ACL).
Ali H ¨urriyeto ˘glu, Erdem Y ¨or¨uk, Deniz Y ¨uret, C ¸ a ˘grı
Yoltar, Burak G ¨urel, Fırat Durus ¸an, Osman Mutlu,
and Arda Akdemir. 2019. Overview of clef 2019
lab protestnews: Extracting protests from news
in a cross-context setting. In Experimental IR
Meets Multilinguality, Multimodality, and Interac-
tion, pages 425–432, Cham. Springer International
Publishing.Ali H ¨urriyeto ˘glu, Vanni Zavarella, Hristo Tanev, Er-
dem Y ¨or¨uk, Ali Safaya, and Osman Mutlu. 2020.
Automated extraction of socio-political events from
news (AESPEN): Workshop and shared task report.
InProceedings of the Workshop on Automated Ex-
traction of Socio-political Events from News 2020 ,
pages 1–6, Marseille, France. European Language
Resources Association (ELRA).
Richard Liaw, Eric Liang, Robert Nishihara, Philipp
Moritz, Joseph E Gonzalez, and Ion Stoica.
2018. Tune: A research platform for distributed
model selection and training. arXiv preprint
arXiv:1807.05118 .
Zachary C Lipton, Charles Elkan, and Balakrishnan
Naryanaswamy. 2014. Optimal thresholding of clas-
sifiers to maximize f1 measure. In Joint European
Conference on Machine Learning and Knowledge
Discovery in Databases , pages 225–239. Springer.
Ilya Loshchilov and Frank Hutter. 2019. Decoupled
weight decay regularization. In International Con-
ference on Learning Representations .
Marius Mosbach, Maksym Andriushchenko, and Diet-
rich Klakow. 2021. On the stability of fine-tuning
fbertg: Misconceptions, explanations, and strong
baselines. In International Conference on Learning
Representations .
Hiroki Nakayama. 2018. seqeval: A python framework
for sequence labeling evaluation. Software available
from https://github.com/chakki-works/seqeval.
Telmo Pires, Eva Schlinger, and Dan Garrette. 2019.
How multilingual is multilingual bert? In Proceed-
ings of the 57th Annual Meeting of the Association
for Computational Linguistics , pages 4996–5001.
Lance Ramshaw and Mitch Marcus. 1995. Text chunk-
ing using transformation-based learning. In Third
Workshop on Very Large Corpora .
Sebastian Ruder. 2021. Recent Advances in Lan-
guage Model Fine-tuning. http://ruder.io/
recent-advances-lm-fine-tuning .
Diana Santos, Nuno Seco, Nuno Cardoso, and Rui
Vilela. 2006. Harem: An advanced ner evalua-
tion contest for portuguese. In quot; In Nicoletta
Calzolari; Khalid Choukri; Aldo Gangemi; Bente
Maegaard; Joseph Mariani; Jan Odjik; Daniel
Tapias (ed) Proceedings of the 5 th International
Conference on Language Resources and Evaluation
(LREC’2006)(Genoa Italy 22-28 May 2006) .
Erik F. Tjong Kim Sang. 2002. Introduction to the
CoNLL-2002 shared task: Language-independent
named entity recognition. In COLING-02: The
6th Conference on Natural Language Learning 2002
(CoNLL-2002) .
Erik F. Tjong Kim Sang and Fien De Meulder.
2003. Introduction to the CoNLL-2003 shared task:
Language-independent named entity recognition. InProceedings of the Seventh Conference on Natu-
ral Language Learning at HLT-NAACL 2003 , pages
142–147.
Thomas Wolf, Lysandre Debut, Victor Sanh, Julien
Chaumond, Clement Delangue, Anthony Moi, Pier-
ric Cistac, Tim Rault, R ´emi Louf, Morgan Funtow-
icz, Joe Davison, Sam Shleifer, Patrick von Platen,
Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu,
Teven Le Scao, Sylvain Gugger, Mariama Drame,
Quentin Lhoest, and Alexander M. Rush. 2020.
Transformers: State-of-the-art natural language pro-
cessing. In Proceedings of the 2020 Conference on
Empirical Methods in Natural Language Processing:
System Demonstrations , pages 38–45, Online. Asso-
ciation for Computational Linguistics.
Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V .
Le, Mohammad Norouzi, Wolfgang Macherey,
Maxim Krikun, Yuan Cao, Qin Gao, Klaus
Macherey, Jeff Klingner, Apurva Shah, Melvin John-
son, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws,
Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith
Stevens, George Kurian, Nishant Patil, Wei Wang,
Cliff Young, Jason Smith, Jason Riesa, Alex Rud-
nick, Oriol Vinyals, Greg Corrado, Macduff Hughes,
and Jeffrey Dean. 2016. Google’s neural machine
translation system: Bridging the gap between human
and machine translation. CoRR , abs/1609.08144.
A Appendix
A.1 Hyper-parameter search
The space search for our hyper-parameter search
was:
•Number of training epochs: value in [20, 25,
30, 40],
•Weight decay: uniform distribution between
0.001 and 1,
•Learning rate: value in [1e-5, 2e-5, 3e-5, 4e-5,
5e-5, 6e-5, 2e-7, 1e-7, 3e-7, 2e-8],
• Adafactor: value in ”True”, ”False”,
•Adam beta 1: uniform distribution between 0
and 1,
•Adam beta 2: uniform distribution between 0
and 1,
•Epsilon: value in [1e-8, 2e-8, 3e-8, 1e-9, 2e-9,
3e-10],
•Maximum gradient norm: uniform distribu-
tion between 0 and 1.For the HyperOpt algorithm we used two set
of hyper-parameters to help finding a good sub-
space. We maximized the macro-F1 on the evalu-
ation dataset, and set the number of initial points
before starting the algorithm to 5.