{ "paper_id": "2019", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:29:55.837375Z" }, "title": "A Multi-task Model for Multilingual Trigger Detection and Classification", "authors": [ { "first": "Sovan", "middle": [], "last": "Kumar", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Saumajit", "middle": [], "last": "Saha", "suffix": "", "affiliation": { "laboratory": "", "institution": "Indian Institute of Technology Patna", "location": {} }, "email": "" }, { "first": "Asif", "middle": [], "last": "Ekbal", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Pushpak", "middle": [], "last": "Bhattacharyya", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "In this paper we present a deep multi-task learning framework for multilingual event and argument trigger detection and classification. In our current work, we identify detection and classification of both event and argument triggers as related tasks and follow a multi-tasking approach to solve them simultaneously in contrast to the previous works where these tasks were solved separately or learning some of the above mentioned tasks jointly. We evaluate the proposed approach with multiple low-resource Indian languages. As there were no datasets available for the Indian languages, we have annotated disaster related news data crawled from the online news portal for different lowresource Indian languages for our experiments. Our empirical evaluation shows that multitask model performs better than the single task model, and classification helps in trigger detection and vice-versa.", "pdf_parse": { "paper_id": "2019", "_pdf_hash": "", "abstract": [ { "text": "In this paper we present a deep multi-task learning framework for multilingual event and argument trigger detection and classification. In our current work, we identify detection and classification of both event and argument triggers as related tasks and follow a multi-tasking approach to solve them simultaneously in contrast to the previous works where these tasks were solved separately or learning some of the above mentioned tasks jointly. We evaluate the proposed approach with multiple low-resource Indian languages. As there were no datasets available for the Indian languages, we have annotated disaster related news data crawled from the online news portal for different lowresource Indian languages for our experiments. Our empirical evaluation shows that multitask model performs better than the single task model, and classification helps in trigger detection and vice-versa.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Event Extraction is an important task in Natural Language Processing (NLP). An event can be an occurrence happening in certain place during a particular interval of time. In text, the word or phrase that describes an event is called event trigger. Argument of an event refers to the attributes such as the location, time of occurrence of the event, participants involved and so on. Therefore event trigger detection, event trigger classification, argument trigger detection and argument trigger classification are the four important sub-tasks of event extraction. In our current paper, we have solved all the four problems using a Multi-task architecture. Multi-task learning (MTL), which essentially means performing more than one related task simultaneously, has been proven to be effective for various NLP tasks in recent times (Ruder, 2017) . The key idea behind MTL is that the inductive transfer of knowledge, learned for a particular task, can help to improve the performance of another task by means of parameter sharing between tasks. According to Caruana (1997) , \"MTL improves generalization by leveraging the domainspecific information contained in the training signals of related tasks\". In our current work, we have identified detection and classification of both event and arguments as two related tasks. As both event and argument trigger detection are sequence labelling problems, we have merged those two sub-tasks into one and used a single loss function. For the same reason, we have merged event and argument trigger classification task into one task and used another loss function. Thus in our proposed architecture, even though we have two main tasks for learning shared representation, we have basically solved four sub-tasks viz. event trigger detection, event trigger classification, argument detection and argument classification. Our proposed architecture has two variants which are further discussed later in this paper. As we are working with low-resource languages which have data sparsity issue, we have proposed a multi-task, multi-lingual architecture which is trained on both Hindi and Bengali data. Due to unavailability of training data in these two languages, we have annotated disaster related news data crawled from online news portals for our experiments.", "cite_spans": [ { "start": 831, "end": 844, "text": "(Ruder, 2017)", "ref_id": "BIBREF30" }, { "start": 1057, "end": 1071, "text": "Caruana (1997)", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Being a very important problem in NLP, Event Extraction has already been explored by the research community for a long time. Some feature based approaches have decomposed the entire event extraction task into two sub-tasks and solved them separately (Ji and Grishman, 2008; Hong et al., 2011; Liao and Grishman, 2010) . But the main problem of this approach is error propagation which is dealt by Riedel and McCallum (2011a) , Riedel and McCallum (2011b) , Li et al. (2013) , Venugopal et al. (2014) using a joint event extraction algorithm. However both of the above approaches have used hand-designed feature. Nguyen and Grishman (2015) propose a Convolutional Neural Network (CNN) for automatic feature extraction. introduce a dynamic multi-pooling CNN which uses a dynamic multi-pooling layer according to event triggers and arguments in multi-event sentences, to capture more crucial information. In another work, Nguyen and Grishman (2016) propose a skip-gram based CNN model which allows nonconsecutive convolution. Ghaeini et al. (2016) propose a forward-backward Recurrent Neural Network (RNN) to detect event triggers which can be in the form of both words or phrases. propose a language independent neural network which uses both CNN and Bi-LSTM for Event detection. Liu et al. (2016) propose to improve the performance of event detection by using the events automatically detected from FrameNet. Though these neural based systems perform well, they still suffer from error propagation issue. To overcome this issue, Nguyen et al. (2016) propose a joint framework with bidirectional RNN. However Liu et al. (2017) observe that joint model achieves insignificant improvements on event detection task. They analyze the problem of joint models on the task of event detection, and propose to use the annotated argument information explicitly for this task. Yang and Mitchell (2016) also propose a joint model for event and entity extraction but in document level instead of sentence level in contrast to most of the previous works. In recent years Liu et al. (2018a) introduce a cross language attention model for event detection where they focus on English and Chinese. Liu et al. (2018b) propose a novel framework to jointly extract multiple event triggers and arguments. Sha et al. (2018) propose a novel dependency bridge RNN which includes syntactic dependency relationships. Dependency relationship is also used by Nguyen and Grishman (2018). They investigate a CNN based on dependency trees to perform event detection. Orr et al. (2018) present a Gated Recurrent Unit (GRU) based model that combines both temporal structure along with syntactic information through an attention mechanism. Event extraction task has also been addressed in specialized tracks dedicated in Text Analysis Conference (TAC). Event extraction in disaster domain in English language is reported in (Tanev et al., 2008; Yun, 2011; Klein et al., 2013; Dittrich and Lucas, 2014; Nugent et al., 2017; Burel et al., 2017) . However, significant attempt to build event extraction system in Indian languages is lacking. In recent times, some of the works are reported in (SharmilaDevi et al., 2017; Sristy et al., 2017; Kuila and Sarkar, 2017; Singh et al., 2017) . To the best of our knowledge, this is the first attempt to solve four important subtasks of event extraction viz. event trigger detection, event trigger classification, argument trigger detection and argument trigger classification simultaneously in a multi-task, multi-lingual setting.", "cite_spans": [ { "start": 250, "end": 273, "text": "(Ji and Grishman, 2008;", "ref_id": "BIBREF12" }, { "start": 274, "end": 292, "text": "Hong et al., 2011;", "ref_id": "BIBREF40" }, { "start": 293, "end": 317, "text": "Liao and Grishman, 2010)", "ref_id": "BIBREF16" }, { "start": 397, "end": 424, "text": "Riedel and McCallum (2011a)", "ref_id": "BIBREF28" }, { "start": 427, "end": 454, "text": "Riedel and McCallum (2011b)", "ref_id": "BIBREF29" }, { "start": 457, "end": 473, "text": "Li et al. (2013)", "ref_id": "BIBREF15" }, { "start": 476, "end": 499, "text": "Venugopal et al. (2014)", "ref_id": "BIBREF38" }, { "start": 1023, "end": 1044, "text": "Ghaeini et al. (2016)", "ref_id": "BIBREF9" }, { "start": 1278, "end": 1295, "text": "Liu et al. (2016)", "ref_id": "BIBREF18" }, { "start": 1607, "end": 1624, "text": "Liu et al. (2017)", "ref_id": "BIBREF19" }, { "start": 1864, "end": 1888, "text": "Yang and Mitchell (2016)", "ref_id": "BIBREF39" }, { "start": 2055, "end": 2073, "text": "Liu et al. (2018a)", "ref_id": "BIBREF17" }, { "start": 2178, "end": 2196, "text": "Liu et al. (2018b)", "ref_id": "BIBREF20" }, { "start": 2281, "end": 2298, "text": "Sha et al. (2018)", "ref_id": "BIBREF32" }, { "start": 2533, "end": 2550, "text": "Orr et al. (2018)", "ref_id": "BIBREF26" }, { "start": 2887, "end": 2907, "text": "(Tanev et al., 2008;", "ref_id": "BIBREF37" }, { "start": 2908, "end": 2918, "text": "Yun, 2011;", "ref_id": "BIBREF40" }, { "start": 2919, "end": 2938, "text": "Klein et al., 2013;", "ref_id": "BIBREF13" }, { "start": 2939, "end": 2964, "text": "Dittrich and Lucas, 2014;", "ref_id": "BIBREF6" }, { "start": 2965, "end": 2985, "text": "Nugent et al., 2017;", "ref_id": "BIBREF25" }, { "start": 2986, "end": 3005, "text": "Burel et al., 2017)", "ref_id": "BIBREF1" }, { "start": 3153, "end": 3180, "text": "(SharmilaDevi et al., 2017;", "ref_id": "BIBREF33" }, { "start": 3181, "end": 3201, "text": "Sristy et al., 2017;", "ref_id": "BIBREF36" }, { "start": 3202, "end": 3225, "text": "Kuila and Sarkar, 2017;", "ref_id": "BIBREF14" }, { "start": 3226, "end": 3245, "text": "Singh et al., 2017)", "ref_id": "BIBREF34" } ], "ref_spans": [], "eq_spans": [], "section": "Related Works", "sec_num": "2" }, { "text": "In this paper, we propose a multi-task, multilingual trigger detection and classification method for Hindi and Bengali in Disaster related news data. For a given Hindi/Bengali sentence, we perform the following tasks simultaneously: (a) Event Trigger Detection: Word or phrase that describes an event is called event trigger. Detecting event triggers is a sequence labeling task. But we formulate our current approach as a multi-class classification task as in Ghaeini et al., 2016) . (b) Event Trigger Classification: Here the task is to classify each event trigger into predefined types. (c) Argument Detection: Arguments are entities, times or values related to an event. Here the task is to detect such trigger words or phrase. (d) Argument Classification: Classify each argument trigger into predefined argument roles. Argument detection is also a sequence labeling task. Like event detection, we also formulate this task as a multi-class classification problem. In most of the previous works, both event and argument detection are considered as two separate tasks. However in our current work, we combine both the tasks into a single task based on our observation. Detailed analysis of news articles reveal the fact that each type of event triggers along with its corresponding arguments follow a particular pattern in a sentence. In the first example, the sentence contains Place argument \u0926 \u0932\u0940 (Delhi) and Time argument \u0936\u093e\u092e 6 \u092c\u091c\u0947 (6pm). Each type of argument is followed by a type specific post-position ('\u092e ' for Place argument and '\u0915\u0947 ' for Time argument). In second example the sentence contains event specific argument like Magnitude (7.2) of earthquake along with Place argument \u0907\u0902 \u0921\u094b \u0928\u0936\u092f\u093e (Indonesia). This type of patterns are often seen in news documents. So it is intuitive to consider both event and argument trigger detection as a single task. For classification also, we merge both the event trigger classification and argument trigger classification as a single task. In this way, we learn all the four above mentioned tasks simultaneously using two loss functions. We perform our experiments using both Hindi and Bengali news datasets in mono-lingual as well as multi-lingual settings. We compare our multi-task learning (MTL) results with single-task learning (STL) results for the above mentioned mono-lingual and multi-lingual settings. For most of the cases we are getting 2% to 7% performance improvement in detection task. However for classification task, we see that the performance improves for some of the classes and for the remaining classes, the model does not perform at par with the other classes. Two contributions of our paper are \u2022 A multi-task, multi-lingual approach for event extraction in Hindi and Bengali for disaster domain. Our proposed system has two variants -(a) The classification output helping in detection (MT1). (b) The detection output helping in classification (MT2). Both the architectures are discussed in methodology section.", "cite_spans": [ { "start": 461, "end": 482, "text": "Ghaeini et al., 2016)", "ref_id": "BIBREF9" } ], "ref_spans": [], "eq_spans": [], "section": "Task Description and Contributions", "sec_num": "3" }, { "text": "\u2022 Provide a benchmark setup for event extraction in Hindi language.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Task Description and Contributions", "sec_num": "3" }, { "text": "The following examples show that each type of event and argument trigger is followed by semantically similar kind of words in a sentence. We highlight the event trigger and different types of argument triggers using different colour codes for better readability. ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Task Description and Contributions", "sec_num": "3" }, { "text": "Our proposed models take sentence of the form [w 0 ,w 1 ,...,w n ] as input. It produces two outputs for two main tasks namely detection (both event and argument) and classification (both event and argument). The detection task predicts the event or argument label (l i ) for each word (w i ) where l i \u2208 I,O,B 1 . As we formulate detection as a multiclass classification task even though it being a sequence labeling task, we use softmax classifier at the final layer. For classification task also, we use softmax classifier at the final layer to classify event and argument trigger into their predefined types. We employ a hard parameter sharing strategy (Caruana, 1993) . We use a shared Bidirectional Long Short-Time Memory (Bi-LSTM) (Schuster and Paliwal, 1997) to capture the contextual information of each word. Figure 1a illustrates the design of first variant of our proposed architecture. Here the classification output of each word is concatenated with the corresponding representation resulting from the shared Bi-LSTM and fed as input to the final detection layer of that word. This is done with the intuition of improving the detection results with the help of classification output. For example if a word is classified as 'None' then it has higher chance of being outside event or argument trigger boundaries. In subsequent sections, we call this architecture as MT1. Figure 1b illustrates the design of second variant of our proposed architecture. Here the detection output of each word is concatenated with the corresponding representation of the shared Bi-LSTM and fed as input to the final classification layer. This is done with the intuition of improving the classification results with the help of detection output.", "cite_spans": [ { "start": 657, "end": 672, "text": "(Caruana, 1993)", "ref_id": "BIBREF3" }, { "start": 738, "end": 766, "text": "(Schuster and Paliwal, 1997)", "ref_id": "BIBREF31" } ], "ref_spans": [ { "start": 819, "end": 828, "text": "Figure 1a", "ref_id": "FIGREF0" }, { "start": 1383, "end": 1392, "text": "Figure 1b", "ref_id": "FIGREF0" } ], "eq_spans": [], "section": "Methodology", "sec_num": "4" }, { "text": "Each word of the input instance is converted to a numeric representation with the help of fast-Text (Grave et al., 2018) word embeddings having dimension 300 (d e ). The pre-trained word vectors are downloaded from fastText website 2 . To learn a mapping between mono-lingual word embeddings and obtain cross-lingual embeddings in order to bridge the language gap between two languages, we use the existing alignment matrices 3 which align monolingual vectors from two lan-2 https://fasttext.cc 3 https://github.com/Babylonpartners/ fastText_multilingual guages in a single vector space (Smith et al., 2017) .", "cite_spans": [ { "start": 100, "end": 120, "text": "(Grave et al., 2018)", "ref_id": "BIBREF10" }, { "start": 587, "end": 607, "text": "(Smith et al., 2017)", "ref_id": "BIBREF35" } ], "ref_spans": [], "eq_spans": [], "section": "Embedding", "sec_num": "4.1" }, { "text": "In order to handle Out-of-Vocabulary (OOV) words in the monolingual setting, we obtain their word embedding vectors from fastText's .bin file. Separate vocabularies for OOV words are created for Hindi and Bengali respectively. We create separate .vec file for the two OOV vocabularies. We similarly transform these vectors of two different languages in a shared space using the existing alignment matrices 3 . It is seen that the performance has significantly improved using crosslingual embeddings for OOV words compared to the method of using zero vectors for representing them. Since there is a lack of annotated data for our task, we create the datasets by crawling online Hindi and Bengali news articles and then annotate them following the TAC KBP 4 guidelines. For annotation, three annotators were employed. We estimate the inter-annotator agreement ratio by ask-ing all the three annotators to annotate 5% of total documents. The multi-rater Kappa (Fleiss, 1971) agreement ratio of 0.82 and 0.85 was observed for Hindi and Bengali news documents respectively.", "cite_spans": [ { "start": 957, "end": 971, "text": "(Fleiss, 1971)", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Embedding", "sec_num": "4.1" }, { "text": "For both the languages, news documents are crawled from online news portal. Every sentence of news documents was pre-processed for four sub-tasks of event extraction viz. event trigger detection, event trigger classification, argument detection and argument classification. Table 2 presents an example of sample annotation. For detection, we use IOB2 1 format (Ramshaw and Marcus, 1999). Our proposed Hindi dataset has two types of disaster events namely natural disaster and man-made disaster which are further classified into twenty seven sub-types. Each event trigger belongs to one of the twenty seven classes, which can be found in Table 8 . Every event has multiple arguments of different roles. Hindi dataset contains eleven types of arguments excluding Type argument type. Bengali dataset also contains eleven type of arguments excluding argument type Intensity. Table 5 contains all the argument types. Some of the argument types common to both Hindi and Bengali, irrespective of the event types, are Place, Time, Casualties and After-effect. Some of the arguments are specific to some particular event types. For example, Magnitude and Epicentre are event specific arguments related to Earthquake. Table 1 presents the dataset statistics for training and the test set of Hindi and Bengali, respectively.", "cite_spans": [], "ref_spans": [ { "start": 274, "end": 281, "text": "Table 2", "ref_id": "TABREF4" }, { "start": 637, "end": 644, "text": "Table 8", "ref_id": "TABREF10" }, { "start": 871, "end": 878, "text": "Table 5", "ref_id": "TABREF7" }, { "start": 1208, "end": 1215, "text": "Table 1", "ref_id": "TABREF2" } ], "eq_spans": [], "section": "Datasets and Experiments", "sec_num": "5" }, { "text": "Optimizer Adam For implementing the deep learning models Python based library Keras (Chollet et al., 2015) with Tensorflow (Abadi et al., 2015) backend is used. All the models are trained for 300 epochs. Training is done using a learning rate of 0.001 and 'Adam' optimizer is used for fast convergence. The data is fed to the neural network in batches of 32. 'Checkpoints' are used to save the best weights of the model based on training accuracy. Table 3 shows the hyper-parameter settings used in the implementation of both the variants of our proposed model. For evaluation precision, recall and F1score are used as the metrics. However in result tables (refer Table 4, Table 5, Table 6, Table 7 and Table 8 ) only F1-score is reported. Table 4, Table 5 and Table 8 show the experimental results for event and argument trigger detection, argument role classification and event trigger classification respectively, where ST denotes Single task, MT1 denotes Multi-task 1, MT2 denotes Multi-task 2 and SP denotes support count. Table 4 shows that multi-task model 1 (MT1) performs well as compared to single task (ST) model for all language settings. For each language setting, performance improvement is maximum in case of I_Event tag. We find that it is 7.3% for Hindi, for Bengali it is 11.5% and for multi-lingual setting it shows improvement of 6.5%. Analyzing the predictions of all the variants of our system reveal that words are usually miss-classified more between the Beginning (B) and Inside (I) tag type of either event or argument instead of events getting miss-classified as argument triggers. Thus we can conclude that the system produces near correct prediction of event and argument trigger in most of the cases, only issue being that it sometimes fail to determine the correct trigger boundary. Figure 2a and Figure 2b show the confusion matrix obtained by MT1 in trigger detection and trigger classification in the multilingual setting.", "cite_spans": [ { "start": 84, "end": 106, "text": "(Chollet et al., 2015)", "ref_id": null }, { "start": 123, "end": 143, "text": "(Abadi et al., 2015)", "ref_id": "BIBREF0" } ], "ref_spans": [ { "start": 448, "end": 455, "text": "Table 3", "ref_id": "TABREF5" }, { "start": 664, "end": 711, "text": "Table 4, Table 5, Table 6, Table 7 and Table 8", "ref_id": "TABREF6" }, { "start": 741, "end": 769, "text": "Table 4, Table 5 and Table 8", "ref_id": "TABREF6" }, { "start": 1029, "end": 1036, "text": "Table 4", "ref_id": "TABREF6" }, { "start": 1815, "end": 1824, "text": "Figure 2a", "ref_id": "FIGREF3" }, { "start": 1829, "end": 1838, "text": "Figure 2b", "ref_id": "FIGREF3" } ], "eq_spans": [], "section": "Loss function for Classification categorical_crossentropy", "sec_num": null }, { "text": "We also perform separate experiments to evaluate our proposed approach with the earlier proposed approaches of separately detecting event and argument triggers from sentences. Table 6 shows the F1-score achieved in event trigger detection and Table 7 shows the F1-score obtained in argument trigger detection for both the Hindi and Bengali datasets. The evaluation shows that there is not any significant loss in performance in simultaneous detection of event and argument triggers compared to individual trigger detection even though there is a marginal improvement in detection of the tag I_Event for Bengali in the argument detection model compared to the model which per- forms simultaneous detection of both triggers.", "cite_spans": [], "ref_spans": [ { "start": 176, "end": 183, "text": "Table 6", "ref_id": "TABREF8" }, { "start": 243, "end": 250, "text": "Table 7", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Comparison With Separate Event and Argument Trigger Detection System", "sec_num": "6.1" }, { "text": "In the following Input Example 1, \u0935\u093e\u0932\u093e\u092e\u0941 \u0916\u0940 \u0935 \u092b\u094b\u091f (jvaalaamukhee visphot\\volcanic erup-tions) is a multi-word event trigger. The tags assigned for this trigger are B_Event and I_Event respectively. In Input Example 2, the event trigger \u0935 \u092b\u094b\u091f (visphot\\eruptions) is tagged as B_Event.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "For the first case, all the variants of the sys-tem predict the event trigger correctly but for the later case, our single task detection system (ST) and multi-task system 2 (MT2) predict it as outside event and argument trigger boundary (O) but multi-task system 1 (MT1) predicts it as inside event trigger (I_Event) rather than beginning of event trigger (B_Event). Thus we can see that all the variants miss-classify the trigger tag with MT1 being able to produce partially correct prediction as it, at least, classifies it to be of event type. However the classification result of the said event trigger in example 2 is correctly predicted by MT1 but it is wrongly predicted by MT2. Here we can see that the classification task is helping in detection task. We provide below a detailed error analysis of the results achieved in classification task (refer to Table 5 and Table 8 ).", "cite_spans": [], "ref_spans": [ { "start": 862, "end": 881, "text": "Table 5 and Table 8", "ref_id": "TABREF7" } ], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "1. In the classification task (refer to Table 5), error analysis reveals that the performance is affected mainly due to two cases : (a) when the Support count of a trigger type is less, (b) when each trigger mention in a sentence is long, i.e. it consists of numerous words. For example, Participant, Time, Place, Casualties and Intensity have better F1-score as the trigger mentions corresponding to these types are in the form of short phrases as well as these types have larger support count. However, roles like After Effect and Reason have comparatively lower performance as these trigger mentions appear in sentences in the form of long phrases. Even though Magnitude has less support count, performance is better compared to the other roles as the trigger mention is in the form of a single word comprising of a numeric figure.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "In Table 8 , we observe the following drawbacks which can possibly lead to erroneous output.", "cite_spans": [], "ref_spans": [ { "start": 3, "end": 10, "text": "Table 8", "ref_id": "TABREF10" } ], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "1. We find that performance decreases for similar types of events. For example, types like Fire, Forest Fire and Industrial Accident are of similar type. We see that the performance of these types is low in Hindi as all of them are present in the dataset, thereby getting missclassified. However in Bengali dataset, we find Fire performs relatively better as there does not exist any sentence having event trigger of type Forest Fire and Industrial Accident.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "2. In Hindi dataset, we find that type Transport Hazard is seen to be misclassified with type Train Collision and type Vehicular Collision, therby leading to poor performance. For Bengali dataset, there hardly exists any trigger of type Train Collision and event trigger of type Vehicular Collision exists in small number. Thus Bengali dataset performs much better for Transport Hazard.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Error Analysis", "sec_num": "6.2" }, { "text": "In this paper, we present a multi-tasking, multilingual architecture for simultaneous detection and classification of event and argument triggers. We have proposed two variants where in each one of them, one task is helping another related task. Our results show that related tasks can definitely share information between them. We also compare our approach with separate models which can be employed for event and argument trigger detection respectively.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and Future Works", "sec_num": "7" }, { "text": "Other future works include developing an endto-end system which will consist of a multi-tasking system such that given a sentence as input, event and argument triggers will be extracted from it and if there exists any link between the extracted event and argument, then the output of the system will be positive and otherwise negative.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and Future Works", "sec_num": "7" }, { "text": "The work reported in this paper is supported by the project titled \"A Platform for Cross-lingual and Multi-lingual Event Monitoring in Indian Languages\", sponsored by IMPRINT-1, Ministry of Human Resource and Development, Government of India. Sovan Kumar Sahoo gratefully acknowledges \"Visvesvaraya PhD Scheme for Electronics and IT\", under the Ministry of Electronics and Information Technology, Government of India.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgement", "sec_num": "8" }, { "text": "The encoding scheme is according to IOB2, where I indicates the tokens that appear within trigger, B denotes the beginning of a trigger and O denotes the outside of an event trigger. The B is used only when two events of the same type appear in consecutive sequence(Ramshaw and Marcus, 1999)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "https://www.nist.gov/tac/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems. Software available from tensorflow.org", "authors": [ { "first": "Mart\u00edn", "middle": [], "last": "Abadi", "suffix": "" }, { "first": "Ashish", "middle": [], "last": "Agarwal", "suffix": "" }, { "first": "Paul", "middle": [], "last": "Barham", "suffix": "" }, { "first": "Eugene", "middle": [], "last": "Brevdo", "suffix": "" }, { "first": "Zhifeng", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Craig", "middle": [], "last": "Citro", "suffix": "" }, { "first": "Greg", "middle": [ "S" ], "last": "Corrado", "suffix": "" }, { "first": "Andy", "middle": [], "last": "Davis", "suffix": "" }, { "first": "Jeffrey", "middle": [], "last": "Dean", "suffix": "" }, { "first": "Matthieu", "middle": [], "last": "Devin", "suffix": "" }, { "first": "Sanjay", "middle": [], "last": "Ghemawat", "suffix": "" }, { "first": "Ian", "middle": [], "last": "Goodfellow", "suffix": "" }, { "first": "Andrew", "middle": [], "last": "Harp", "suffix": "" }, { "first": "Geoffrey", "middle": [], "last": "Irving", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Isard", "suffix": "" }, { "first": "Yangqing", "middle": [], "last": "Jia", "suffix": "" }, { "first": "Rafal", "middle": [], "last": "Jozefowicz", "suffix": "" }, { "first": "Lukasz", "middle": [], "last": "Kaiser", "suffix": "" }, { "first": "Manjunath", "middle": [], "last": "Kudlur", "suffix": "" }, { "first": "Josh", "middle": [], "last": "Levenberg", "suffix": "" } ], "year": 2015, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mart\u00edn Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Cor- rado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dandelion Man\u00e9, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schus- ter, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Vi\u00e9gas, Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. 2015. TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems. Software available from tensorflow.org.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "On semantics and deep learning for event detection in crisis situations", "authors": [ { "first": "Gr\u00e9goire", "middle": [], "last": "Burel", "suffix": "" }, { "first": "Hassan", "middle": [], "last": "Saif", "suffix": "" }, { "first": "Miriam", "middle": [], "last": "Fernandez", "suffix": "" }, { "first": "Harith", "middle": [], "last": "Alani", "suffix": "" } ], "year": 2017, "venue": "Workshop on Semantic Deep Learning (SemDeep)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Gr\u00e9goire Burel, Hassan Saif, Miriam Fernandez, and Harith Alani. 2017. On semantics and deep learning for event detection in crisis situations. Workshop on Semantic Deep Learning (SemDeep), at ESWC 2017, 29 May 2017, Portoroz, Slovenia.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Multitask learning. Machine learning", "authors": [ { "first": "Rich", "middle": [], "last": "Caruana", "suffix": "" } ], "year": 1997, "venue": "", "volume": "28", "issue": "", "pages": "41--75", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rich Caruana. 1997. Multitask learning. Machine learning, 28(1):41-75.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Multitask learning: A knowledge-based source of inductive bias", "authors": [ { "first": "Richard", "middle": [], "last": "Caruana", "suffix": "" } ], "year": 1993, "venue": "Proceedings of the Tenth International Conference on Machine Learning", "volume": "", "issue": "", "pages": "41--48", "other_ids": {}, "num": null, "urls": [], "raw_text": "Richard Caruana. 1993. Multitask learning: A knowledge-based source of inductive bias. In Pro- ceedings of the Tenth International Conference on Machine Learning, pages 41-48. Morgan Kauf- mann.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Event extraction via dynamic multipooling convolutional neural networks", "authors": [ { "first": "Yubo", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Liheng", "middle": [], "last": "Xu", "suffix": "" }, { "first": "Kang", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Daojian", "middle": [], "last": "Zeng", "suffix": "" }, { "first": "Jun", "middle": [], "last": "Zhao", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing", "volume": "1", "issue": "", "pages": "167--176", "other_ids": {}, "num": null, "urls": [], "raw_text": "Yubo Chen, Liheng Xu, Kang Liu, Daojian Zeng, and Jun Zhao. 2015. Event extraction via dynamic multi- pooling convolutional neural networks. In Proceed- ings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th Interna- tional Joint Conference on Natural Language Pro- cessing (Volume 1: Long Papers), volume 1, pages 167-176.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Is this twitter event a disaster?", "authors": [ { "first": "Andr\u00e9", "middle": [], "last": "Dittrich", "suffix": "" }, { "first": "Christian", "middle": [], "last": "Lucas", "suffix": "" } ], "year": 2014, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Andr\u00e9 Dittrich and Christian Lucas. 2014. Is this twitter event a disaster? AGILE Digital Editions.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "A language-independent neural network for event detection", "authors": [ { "first": "Xiaocheng", "middle": [], "last": "Feng", "suffix": "" }, { "first": "Bing", "middle": [], "last": "Qin", "suffix": "" }, { "first": "Ting", "middle": [], "last": "Liu", "suffix": "" } ], "year": 2018, "venue": "Science China Information Sciences", "volume": "61", "issue": "9", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Xiaocheng Feng, Bing Qin, and Ting Liu. 2018. A language-independent neural network for event detection. Science China Information Sciences, 61(9):092106.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Measuring nominal scale agreement among many raters", "authors": [ { "first": "L", "middle": [], "last": "Joseph", "suffix": "" }, { "first": "", "middle": [], "last": "Fleiss", "suffix": "" } ], "year": 1971, "venue": "Psychological bulletin", "volume": "76", "issue": "5", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Joseph L Fleiss. 1971. Measuring nominal scale agree- ment among many raters. Psychological bulletin, 76(5):378.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Event nugget detection with forward-backward recurrent neural networks", "authors": [ { "first": "Reza", "middle": [], "last": "Ghaeini", "suffix": "" }, { "first": "Xiaoli", "middle": [], "last": "Fern", "suffix": "" }, { "first": "Liang", "middle": [], "last": "Huang", "suffix": "" }, { "first": "Prasad", "middle": [], "last": "Tadepalli", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics", "volume": "2", "issue": "", "pages": "369--373", "other_ids": {}, "num": null, "urls": [], "raw_text": "Reza Ghaeini, Xiaoli Fern, Liang Huang, and Prasad Tadepalli. 2016. Event nugget detection with forward-backward recurrent neural networks. In Proceedings of the 54th Annual Meeting of the As- sociation for Computational Linguistics (Volume 2: Short Papers), volume 2, pages 369-373.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Learning Word Vectors for 157 Languages", "authors": [ { "first": "Edouard", "middle": [], "last": "Grave", "suffix": "" }, { "first": "Piotr", "middle": [], "last": "Bojanowski", "suffix": "" }, { "first": "Prakhar", "middle": [], "last": "Gupta", "suffix": "" }, { "first": "Armand", "middle": [], "last": "Joulin", "suffix": "" }, { "first": "Tomas", "middle": [], "last": "Mikolov", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the International Conference on Language Resources and Evaluation (LREC 2018)", "volume": "", "issue": "", "pages": "3483--3487", "other_ids": {}, "num": null, "urls": [], "raw_text": "Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Ar- mand Joulin, and Tomas Mikolov. 2018. Learning Word Vectors for 157 Languages. In Proceedings of the International Conference on Language Re- sources and Evaluation (LREC 2018), pages 3483- 3487.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Using cross-entity inference to improve event extraction", "authors": [ { "first": "Yu", "middle": [], "last": "Hong", "suffix": "" }, { "first": "Jianfeng", "middle": [], "last": "Zhang", "suffix": "" }, { "first": "Bin", "middle": [], "last": "Ma", "suffix": "" }, { "first": "Jianmin", "middle": [], "last": "Yao", "suffix": "" }, { "first": "Guodong", "middle": [], "last": "Zhou", "suffix": "" }, { "first": "Qiaoming", "middle": [], "last": "Zhu", "suffix": "" } ], "year": 2011, "venue": "Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies", "volume": "1", "issue": "", "pages": "1127--1136", "other_ids": {}, "num": null, "urls": [], "raw_text": "Yu Hong, Jianfeng Zhang, Bin Ma, Jianmin Yao, Guodong Zhou, and Qiaoming Zhu. 2011. Us- ing cross-entity inference to improve event extrac- tion. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Hu- man Language Technologies-Volume 1, pages 1127- 1136. Association for Computational Linguistics.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Refining event extraction through cross-document inference", "authors": [ { "first": "Heng", "middle": [], "last": "Ji", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2008, "venue": "Proceedings of ACL-08: HLT", "volume": "", "issue": "", "pages": "254--262", "other_ids": {}, "num": null, "urls": [], "raw_text": "Heng Ji and Ralph Grishman. 2008. Refining event extraction through cross-document inference. Pro- ceedings of ACL-08: HLT, pages 254-262.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Emergency event detection in twitter streams based on natural language processing", "authors": [ { "first": "Bernhard", "middle": [], "last": "Klein", "suffix": "" }, { "first": "Federico", "middle": [], "last": "Castanedo", "suffix": "" }, { "first": "Inigo", "middle": [], "last": "Elejalde", "suffix": "" }, { "first": "Diego", "middle": [], "last": "L\u00f3pez-De Ipina", "suffix": "" }, { "first": "Alejandro", "middle": [ "Prada" ], "last": "Nespral", "suffix": "" } ], "year": 2013, "venue": "International Conference on Ubiquitous Computing and Ambient Intelligence", "volume": "", "issue": "", "pages": "239--246", "other_ids": {}, "num": null, "urls": [], "raw_text": "Bernhard Klein, Federico Castanedo, Inigo Elejalde, Diego L\u00f3pez-de Ipina, and Alejandro Prada Nespral. 2013. Emergency event detection in twitter streams based on natural language processing. In Inter- national Conference on Ubiquitous Computing and Ambient Intelligence, pages 239-246. Springer.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "An event extraction system via neural networks. FIRE (Working Notes)", "authors": [ { "first": "Alapan", "middle": [], "last": "Kuila", "suffix": "" }, { "first": "Sudeshna", "middle": [], "last": "Sarkar", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "136--139", "other_ids": {}, "num": null, "urls": [], "raw_text": "Alapan Kuila and Sudeshna Sarkar. 2017. An event ex- traction system via neural networks. FIRE (Working Notes), pages 136-139.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Joint event extraction via structured prediction with global features", "authors": [ { "first": "Qi", "middle": [], "last": "Li", "suffix": "" }, { "first": "Ji", "middle": [], "last": "Heng", "suffix": "" }, { "first": "Liang", "middle": [], "last": "Huang", "suffix": "" } ], "year": 2013, "venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics", "volume": "1", "issue": "", "pages": "73--82", "other_ids": {}, "num": null, "urls": [], "raw_text": "Qi Li, Heng Ji, and Liang Huang. 2013. Joint event extraction via structured prediction with global fea- tures. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Vol- ume 1: Long Papers), volume 1, pages 73-82.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Using document level cross-event inference to improve event extraction", "authors": [ { "first": "Shasha", "middle": [], "last": "Liao", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2010, "venue": "Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "789--797", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shasha Liao and Ralph Grishman. 2010. Using doc- ument level cross-event inference to improve event extraction. In Proceedings of the 48th Annual Meet- ing of the Association for Computational Linguistics, pages 789-797. Association for Computational Lin- guistics.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Event detection via gated multilingual attention mechanism", "authors": [ { "first": "Jian", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Yubo", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Kang", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Jun", "middle": [], "last": "Zhao", "suffix": "" } ], "year": 2018, "venue": "Thirty-Second AAAI Conference on Artificial Intelligence", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jian Liu, Yubo Chen, Kang Liu, and Jun Zhao. 2018a. Event detection via gated multilingual attention mechanism. In Thirty-Second AAAI Conference on Artificial Intelligence.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Leveraging framenet to improve automatic event detection", "authors": [ { "first": "Shulin", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Yubo", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Shizhu", "middle": [], "last": "He", "suffix": "" }, { "first": "Kang", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Jun", "middle": [], "last": "Zhao", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics", "volume": "1", "issue": "", "pages": "2134--2143", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shulin Liu, Yubo Chen, Shizhu He, Kang Liu, and Jun Zhao. 2016. Leveraging framenet to improve automatic event detection. In Proceedings of the 54th Annual Meeting of the Association for Compu- tational Linguistics (Volume 1: Long Papers), vol- ume 1, pages 2134-2143.", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Exploiting argument information to improve event detection via supervised attention mechanisms", "authors": [ { "first": "Shulin", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Yubo", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Kang", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Jun", "middle": [], "last": "Zhao", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shulin Liu, Yubo Chen, Kang Liu, Jun Zhao, et al. 2017. Exploiting argument information to improve event detection via supervised attention mechanisms.", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "Jointly multiple events extraction via attention-based graph information aggregation", "authors": [ { "first": "Xiao", "middle": [], "last": "Liu", "suffix": "" }, { "first": "Zhunchen", "middle": [], "last": "Luo", "suffix": "" }, { "first": "Heyan", "middle": [], "last": "Huang", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "1247--1256", "other_ids": {}, "num": null, "urls": [], "raw_text": "Xiao Liu, Zhunchen Luo, and Heyan Huang. 2018b. Jointly multiple events extraction via attention-based graph information aggregation. In Proceedings of the 2018 Conference on Empirical Methods in Nat- ural Language Processing, pages 1247-1256.", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "Joint event extraction via recurrent neural networks", "authors": [ { "first": "Kyunghyun", "middle": [], "last": "Thien Huu Nguyen", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Cho", "suffix": "" }, { "first": "", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "", "issue": "", "pages": "300--309", "other_ids": {}, "num": null, "urls": [], "raw_text": "Thien Huu Nguyen, Kyunghyun Cho, and Ralph Grish- man. 2016. Joint event extraction via recurrent neu- ral networks. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies, pages 300-309.", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "Event detection and domain adaptation with convolutional neural networks", "authors": [ { "first": "Huu", "middle": [], "last": "Thien", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Nguyen", "suffix": "" }, { "first": "", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing", "volume": "2", "issue": "", "pages": "365--371", "other_ids": {}, "num": null, "urls": [], "raw_text": "Thien Huu Nguyen and Ralph Grishman. 2015. Event detection and domain adaptation with convolutional neural networks. In Proceedings of the 53rd Annual Meeting of the Association for Computational Lin- guistics and the 7th International Joint Conference on Natural Language Processing (Volume 2: Short Papers), volume 2, pages 365-371.", "links": null }, "BIBREF23": { "ref_id": "b23", "title": "Modeling skip-grams for event detection with convolutional neural networks", "authors": [ { "first": "Huu", "middle": [], "last": "Thien", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Nguyen", "suffix": "" }, { "first": "", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "886--891", "other_ids": {}, "num": null, "urls": [], "raw_text": "Thien Huu Nguyen and Ralph Grishman. 2016. Mod- eling skip-grams for event detection with convolu- tional neural networks. In Proceedings of the 2016 Conference on Empirical Methods in Natural Lan- guage Processing, pages 886-891.", "links": null }, "BIBREF24": { "ref_id": "b24", "title": "Graph convolutional networks with argument-aware pooling for event detection", "authors": [ { "first": "Huu", "middle": [], "last": "Thien", "suffix": "" }, { "first": "Ralph", "middle": [], "last": "Nguyen", "suffix": "" }, { "first": "", "middle": [], "last": "Grishman", "suffix": "" } ], "year": 2018, "venue": "Thirty-Second AAAI Conference on Artificial Intelligence", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Thien Huu Nguyen and Ralph Grishman. 2018. Graph convolutional networks with argument-aware pool- ing for event detection. In Thirty-Second AAAI Con- ference on Artificial Intelligence.", "links": null }, "BIBREF25": { "ref_id": "b25", "title": "A comparison of classification models for natural disaster and critical event detection from news", "authors": [ { "first": "Tim", "middle": [], "last": "Nugent", "suffix": "" }, { "first": "Fabio", "middle": [], "last": "Petroni", "suffix": "" }, { "first": "Natraj", "middle": [], "last": "Raman", "suffix": "" }, { "first": "Lucas", "middle": [], "last": "Carstens", "suffix": "" }, { "first": "Jochen L", "middle": [], "last": "Leidner", "suffix": "" } ], "year": 2017, "venue": "2017 IEEE International Conference on", "volume": "", "issue": "", "pages": "3750--3759", "other_ids": {}, "num": null, "urls": [], "raw_text": "Tim Nugent, Fabio Petroni, Natraj Raman, Lucas Carstens, and Jochen L Leidner. 2017. A compar- ison of classification models for natural disaster and critical event detection from news. In Big Data (Big Data), 2017 IEEE International Conference on, pages 3750-3759. IEEE.", "links": null }, "BIBREF26": { "ref_id": "b26", "title": "Event detection with neural networks: A rigorous empirical evaluation", "authors": [ { "first": "Walker", "middle": [], "last": "Orr", "suffix": "" }, { "first": "Prasad", "middle": [], "last": "Tadepalli", "suffix": "" }, { "first": "Xiaoli", "middle": [], "last": "Fern", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "999--1004", "other_ids": {}, "num": null, "urls": [], "raw_text": "Walker Orr, Prasad Tadepalli, and Xiaoli Fern. 2018. Event detection with neural networks: A rigorous empirical evaluation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Lan- guage Processing, pages 999-1004.", "links": null }, "BIBREF27": { "ref_id": "b27", "title": "Text chunking using transformation-based learning", "authors": [ { "first": "A", "middle": [], "last": "Lance", "suffix": "" }, { "first": "Mitchell P", "middle": [], "last": "Ramshaw", "suffix": "" }, { "first": "", "middle": [], "last": "Marcus", "suffix": "" } ], "year": 1999, "venue": "Natural language processing using very large corpora", "volume": "", "issue": "", "pages": "157--176", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lance A Ramshaw and Mitchell P Marcus. 1999. Text chunking using transformation-based learning. In Natural language processing using very large cor- pora, pages 157-176. Springer.", "links": null }, "BIBREF28": { "ref_id": "b28", "title": "Fast and robust joint models for biomedical event extraction", "authors": [ { "first": "Sebastian", "middle": [], "last": "Riedel", "suffix": "" }, { "first": "Andrew", "middle": [], "last": "Mccallum", "suffix": "" } ], "year": 2011, "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "1--12", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sebastian Riedel and Andrew McCallum. 2011a. Fast and robust joint models for biomedical event extrac- tion. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 1- 12. Association for Computational Linguistics.", "links": null }, "BIBREF29": { "ref_id": "b29", "title": "Robust biomedical event extraction with dual decomposition and minimal domain adaptation", "authors": [ { "first": "Sebastian", "middle": [], "last": "Riedel", "suffix": "" }, { "first": "Andrew", "middle": [], "last": "Mccallum", "suffix": "" } ], "year": 2011, "venue": "Proceedings of the BioNLP Shared Task 2011 Workshop", "volume": "", "issue": "", "pages": "46--50", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sebastian Riedel and Andrew McCallum. 2011b. Ro- bust biomedical event extraction with dual decom- position and minimal domain adaptation. In Pro- ceedings of the BioNLP Shared Task 2011 Workshop, pages 46-50. Association for Computational Lin- guistics.", "links": null }, "BIBREF30": { "ref_id": "b30", "title": "An overview of multi-task learning in", "authors": [ { "first": "Sebastian", "middle": [], "last": "Ruder", "suffix": "" } ], "year": 2017, "venue": "deep neural networks", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1706.05098" ] }, "num": null, "urls": [], "raw_text": "Sebastian Ruder. 2017. An overview of multi-task learning in deep neural networks. arXiv preprint arXiv:1706.05098.", "links": null }, "BIBREF31": { "ref_id": "b31", "title": "Bidirectional recurrent neural networks", "authors": [ { "first": "Mike", "middle": [], "last": "Schuster", "suffix": "" }, { "first": "K", "middle": [], "last": "Kuldip", "suffix": "" }, { "first": "", "middle": [], "last": "Paliwal", "suffix": "" } ], "year": 1997, "venue": "IEEE Transactions on Signal Processing", "volume": "45", "issue": "11", "pages": "2673--2681", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mike Schuster and Kuldip K Paliwal. 1997. Bidirec- tional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11):2673-2681.", "links": null }, "BIBREF32": { "ref_id": "b32", "title": "Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction", "authors": [ { "first": "Lei", "middle": [], "last": "Sha", "suffix": "" }, { "first": "Feng", "middle": [], "last": "Qian", "suffix": "" }, { "first": "Baobao", "middle": [], "last": "Chang", "suffix": "" }, { "first": "Zhifang", "middle": [], "last": "Sui", "suffix": "" } ], "year": 2018, "venue": "Thirty-Second AAAI Conference on Artificial Intelligence", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lei Sha, Feng Qian, Baobao Chang, and Zhifang Sui. 2018. Jointly extracting event triggers and argu- ments by dependency-bridge rnn and tensor-based argument interaction. In Thirty-Second AAAI Con- ference on Artificial Intelligence.", "links": null }, "BIBREF33": { "ref_id": "b33", "title": "Kce_dalab@ eventxtract-il-fire2017: Event extraction using support vector machines. FIRE (Working Notes)", "authors": [ { "first": "V", "middle": [], "last": "Sharmiladevi", "suffix": "" }, { "first": "G", "middle": [], "last": "Kannimuthu", "suffix": "" }, { "first": "", "middle": [], "last": "Safeeq", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "144--146", "other_ids": {}, "num": null, "urls": [], "raw_text": "V SharmilaDevi, S Kannimuthu, and G Safeeq. 2017. Kce_dalab@ eventxtract-il-fire2017: Event extrac- tion using support vector machines. FIRE (Working Notes), pages 144-146.", "links": null }, "BIBREF34": { "ref_id": "b34", "title": "Event classification and location prediction from tweets during disasters", "authors": [ { "first": "Jyoti", "middle": [ "Prakash" ], "last": "Singh", "suffix": "" }, { "first": "K", "middle": [], "last": "Yogesh", "suffix": "" }, { "first": "", "middle": [], "last": "Dwivedi", "suffix": "" }, { "first": "P", "middle": [], "last": "Nripendra", "suffix": "" }, { "first": "Abhinav", "middle": [], "last": "Rana", "suffix": "" }, { "first": "Kawaljeet Kaur", "middle": [], "last": "Kumar", "suffix": "" }, { "first": "", "middle": [], "last": "Kapoor", "suffix": "" } ], "year": 2017, "venue": "Annals of Operations Research", "volume": "", "issue": "", "pages": "1--21", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jyoti Prakash Singh, Yogesh K Dwivedi, Nripendra P Rana, Abhinav Kumar, and Kawaljeet Kaur Kapoor. 2017. Event classification and location prediction from tweets during disasters. Annals of Operations Research, pages 1-21.", "links": null }, "BIBREF35": { "ref_id": "b35", "title": "Offline bilingual word vectors, orthogonal transformations and the inverted softmax", "authors": [ { "first": "L", "middle": [], "last": "Samuel", "suffix": "" }, { "first": "", "middle": [], "last": "Smith", "suffix": "" }, { "first": "H", "middle": [ "P" ], "last": "David", "suffix": "" }, { "first": "Steven", "middle": [], "last": "Turban", "suffix": "" }, { "first": "Nils", "middle": [ "Y" ], "last": "Hamblin", "suffix": "" }, { "first": "", "middle": [], "last": "Hammerla", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1702.03859" ] }, "num": null, "urls": [], "raw_text": "Samuel L Smith, David HP Turban, Steven Hamblin, and Nils Y Hammerla. 2017. Offline bilingual word vectors, orthogonal transformations and the inverted softmax. arXiv preprint arXiv:1702.03859.", "links": null }, "BIBREF36": { "ref_id": "b36", "title": "Event extraction from social media text using conditional random fields", "authors": [ { "first": "", "middle": [], "last": "Nagesh Bhattu Sristy", "suffix": "" }, { "first": "", "middle": [], "last": "Satya Krishna", "suffix": "" }, { "first": "", "middle": [], "last": "Durvasula", "suffix": "" }, { "first": "", "middle": [], "last": "Somayajulu", "suffix": "" } ], "year": 2017, "venue": "FIRE (Working Notes)", "volume": "", "issue": "", "pages": "140--143", "other_ids": {}, "num": null, "urls": [], "raw_text": "Nagesh Bhattu Sristy, N Satya Krishna, and Durva- sula VLN Somayajulu. 2017. Event extraction from social media text using conditional random fields. In FIRE (Working Notes), pages 140-143.", "links": null }, "BIBREF37": { "ref_id": "b37", "title": "Real-time news event extraction for global crisis monitoring", "authors": [ { "first": "Hristo", "middle": [], "last": "Tanev", "suffix": "" }, { "first": "Jakub", "middle": [], "last": "Piskorski", "suffix": "" }, { "first": "Martin", "middle": [], "last": "Atkinson", "suffix": "" } ], "year": 2008, "venue": "International Conference on Application of Natural Language to Information Systems", "volume": "", "issue": "", "pages": "207--218", "other_ids": {}, "num": null, "urls": [], "raw_text": "Hristo Tanev, Jakub Piskorski, and Martin Atkinson. 2008. Real-time news event extraction for global crisis monitoring. In International Conference on Application of Natural Language to Information Sys- tems, pages 207-218. Springer.", "links": null }, "BIBREF38": { "ref_id": "b38", "title": "Relieving the computational bottleneck: Joint inference for event extraction with high-dimensional features", "authors": [ { "first": "Deepak", "middle": [], "last": "Venugopal", "suffix": "" }, { "first": "Chen", "middle": [], "last": "Chen", "suffix": "" }, { "first": "Vibhav", "middle": [], "last": "Gogate", "suffix": "" }, { "first": "Vincent", "middle": [], "last": "Ng", "suffix": "" } ], "year": 2014, "venue": "Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)", "volume": "", "issue": "", "pages": "831--843", "other_ids": {}, "num": null, "urls": [], "raw_text": "Deepak Venugopal, Chen Chen, Vibhav Gogate, and Vincent Ng. 2014. Relieving the computational bot- tleneck: Joint inference for event extraction with high-dimensional features. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 831-843.", "links": null }, "BIBREF39": { "ref_id": "b39", "title": "Joint extraction of events and entities within a document context", "authors": [ { "first": "Bishan", "middle": [], "last": "Yang", "suffix": "" }, { "first": "Tom", "middle": [], "last": "Mitchell", "suffix": "" } ], "year": 2016, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1609.03632" ] }, "num": null, "urls": [], "raw_text": "Bishan Yang and Tom Mitchell. 2016. Joint extrac- tion of events and entities within a document context. arXiv preprint arXiv:1609.03632.", "links": null }, "BIBREF40": { "ref_id": "b40", "title": "Disaster events detection using twitter data", "authors": [ { "first": "Hong-Won", "middle": [], "last": "Yun", "suffix": "" } ], "year": 2011, "venue": "Journal of information and communication convergence engineering", "volume": "9", "issue": "1", "pages": "69--73", "other_ids": {}, "num": null, "urls": [], "raw_text": "Hong-Won Yun. 2011. Disaster events detection using twitter data. Journal of information and communi- cation convergence engineering, 9(1):69-73.", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "text": "Architecture of Our Proposed Models", "uris": null, "num": null }, "FIGREF3": { "type_str": "figure", "text": "Confusion Matrix : MT1 in Multilingual Setting.", "uris": null, "num": null }, "TABREF2": { "type_str": "table", "html": null, "text": "Dataset Statistics", "content": "", "num": null }, "TABREF4": { "type_str": "table", "html": null, "text": "Sample annotation for the sentence given in Example-2 in Task Description and Contribution Section", "content": "
", "num": null }, "TABREF5": { "type_str": "table", "html": null, "text": "Hyper-parameter Settings", "content": "
", "num": null }, "TABREF6": { "type_str": "table", "html": null, "text": "Trigger Detection (Events and Arguments) Results", "content": "
HindiBengaliMulti-Lingual
ST MT1 MT2SPST MT1 MT2SPST MT1 MT2SP
Participant 0.35 0.42 0.38539 0.43 0.43 0.41816 0.36 0.41 0.36 1355
Epicentre0.59 0.46 0.29220.48 0.27 0.46490.40.20.3571
After Effect 0.30.35 0.31 2828 0.36 0.36 0.35 1648 0.32 0.31 0.33 4476
Reason0.140.10.12354 0.26 0.21 0.20280 0.16 0.16 0.18634
Magnitude 0.560.60.62400.52 0.51 0.44250.47 0.56 0.5465
Place0.57 0.58 0.56 2369 0.61 0.59 0.61 1588 0.58 0.57 0.56 3957
Casualties 0.58 0.59 0.58 1969 0.73 0.73 0.72 2578 0.65 0.66 0.65 4547
Name0.26 0.32 0.276700090.250.30.2376
Type----0.20 0.20 0.24290.19 0.11 0.3729
Intensity0.54 0.440.4191----0.45 0.33 0.27191
Time0.65 0.66 0.63804 0.84 0.85 0.84 2029 0.79 0.77 0.78 2833
Speed0.18 0.110170.36 0.31 0.4640.19 0.36 0.2721
", "num": null }, "TABREF7": { "type_str": "table", "html": null, "text": "", "content": "", "num": null }, "TABREF8": { "type_str": "table", "html": null, "text": "Result of Event Trigger Detection as Only Task.", "content": "
Hindi Bengali
B-Arg 0.490.57
I-Arg0.490.64
", "num": null }, "TABREF9": { "type_str": "table", "html": null, "text": "Result of Argument Detection as Only Task.", "content": "
HindiBengaliMulti-Lingual
ST MT1 MT2 SPST MT1 MT2 SPST MT1 MT2 SP
Armed Conflicts0.20.40.3170.22 0.16 0.22 126 0.21 0.19 0.24 133
Avalanches 0.57 0.61 0.6230----0.51 0.57 0.5730
Aviation Hazard0.35 0.43 0.4643 0.56 0.47 0.3434 0.48 0.34 0.4177
Blizzard0.490.60.511900070.44 0.410.626
Cold Wave 0.53 0.48 0.5326 0.50 0.50 0.5040.52 0.45 0.4930
Cyclone0.40.49 0.3620----0.51 0.45 0.4520
Earthquake 0.69 0.73 0.66 115 0.75 0.74 0.6887 0.71 0.63 0.71 202
Epidemic----0.33 0.33 0.3361 0.340.30.361
Fire0.27 0.26 0.25 114 0.68 0.68 0.66 120 0.44 0.45 0.48 234
Floods0.560.60.727 0.40 0.67 0.5010.64 0.77 0.6628
Forest Fire 0.32 0.31 0.2963----0.330.30.2463
Hail Storms 0.41 0.46 0.3941----0.45 0.52 0.4641
Heat Wave 0.39 0.48 0.3966 0.33 0.24 0.4390.36 0.37 0.4175
Hurricane 0.530.60.3835----0.48 0.47 0.4535
Industrial Accident0.21 0.21 0.17 11300.25030.17 0.18 0.15 116
Landslide0.43 0.38 0.4469 0.74 0.71 0.5990.470.50.4678
Normal Bombing0.180.20.2290.61 0.62 0.58 292 0.57 0.55 0.56 301
Pandemic----0.26 0.23 0.2587 0.17 0.29 0.3287
Riots0.29 0.38 0.3132 0.26 0.31 0.2344 0.280.20.2476
Shootout0.49 0.49 0.44 110 0.56 0.54 0.52 177 0.51 0.520.5287
Storm0.20.22 0.2924 0.45 0.42 0.4226 0.43 0.32 0.3450
Suicide Attack0.64 0.64 0.68 154 0.57 0.62 0.56 123 0.60.59 0.58 277
Surgical Strikes00020.40 0.36 0.4464 0.41 0.38 0.3666
Terrorist Attack0.61 0.61 0.6295 0.32 0.37 0.34 147 0.47 0.48 0.49 242
Tornado0.43 0.49 0.3532 0.570.40.5740.43 0.38 0.4336
Train Collision0.52 0.44 0.537200010.460.40.573
Transport Hazards0.13 0.18 0.1879 0.49 0.47 0.43 127 0.40.36 0.37 206
Tsunami----0.17 0.17 0.1710 0.32 0.13 0.12 0.32
Vehicular Collision0.56 0.52 0.4993 0.43 0.45 0.4839 0.44 0.48 0.46 132
Volcano0.50.42 0.5233----0.48 0.45 0.4333
", "num": null }, "TABREF10": { "type_str": "table", "html": null, "text": "Event Trigger Classification Results", "content": "", "num": null } } } }