Acronym Extraction and Acronym Disambiguation Shared Tasks at the Scientific Document Understanding Workshop 2022 Amir Pouran Ben Veyseh1 , Nicole Meister2 , Franck Dernoncourt3 and Thien Huu Nguyen1 1 Department of Computer and Information Science, University of Oregon 2 Department of Electrical and Computer Engineering, Princeton University 3 Adobe Research Abstract Acronyms are short forms of longer phrases that facilitate the communication, specifically in technical domain that are replete with lengthy phrases. Due to the prevalence of acronyms in various types of documents, it is useful for document understanding systems to have the capability of correctly processing acronyms in text. More specifically, a system should be capable of recognizing the acronym and their long-forms in text (i.e., acronym extraction) and also to provide the correct meaning for the acronyms in case their long-form is missing from the document (i.e., acronym disambiguation). Due to their importance, both acronym extraction (AE) and acronym disambiguation (AD) are studied in the literature. However, the prior works are limited to English and specific domains (e.g., biomedical). To address this limitations, we introduce new resources for AE and AD in multiple languages and domains. Moreover, we organized two shared tasks on multilingual and multi-domain AE and AD. This paper gives an overview of the proposed resources and the participating systems in both shared tasks. Keywords Acronym Extraction, Acronym Disambiguation, Multi-lingual, Scientific Document Understanding 1. Introduction traction [1, 2] and question answering [3, 4]. Beyond AE, other related work looked at definition extraction Technical documents are normally replete with domain- [5, 6, 7, 8, 9] and mathematical symbol definition [10]. specific phrases that might be lengthy to repeat in every An automatic acronym understanding system should mention. As such, to facilitate communication, acronyms be able to recognize the mentions of the acronyms and are heavily employed in technical writing. Concretely, their meanings in text. This task is called Acronym Ex- an acronym is defined as a shortened form of a longer traction (AE). For instance, in the sentence “All input fea- phrases and consists of few letters selected from the long tures are encoded by the Long Short-Term Memory (LSTM) phrase. Using acronyms saves space and could help the network”, an acronym, i.e., “LSTM”, and a long-form, i.e., audience to more easily read the documents. However, “Long Short-Term Memory”, are provided. An AE sys- they might also propose challenges for those that are not tem should be able to recognize the acronym and the familiar with the meaning of the acronym. The acronyms long-form in the sentence. This task is normally mod- that are not defined in a technical document prevent the eled as a sequence classification. In particular, the in- efficient communication of concepts due to lack of clar- put sentence is sent to a sequential model (e.g., Recur- ity. Therefore, providing the meaning for acronyms is rent Neural Network (RNN)) to predict the boundaries an important requirement for any technical document for the acronym and the long-form. Another task that to avoid any confusion about the concepts mentioned in an automatic acronym understanding system should be the document. Manual glossaries could be an option to capable is acronym disambiguation (AD). In this task, address this limitation. However, they might not be com- the goal is to provide the correct meaning for a given plete and also preparing them takes considerable amount acronym in a sentence or paragraph while its long-form of time in case the number of acronyms in the document is missing from the context. For instance, in the sen- are huge. Thus, automatic processing of acronyms is tence “The event is fully covered by CNN ”, the meaning highly demanded to facilitate writing and reading tech- of the acronym “CNN ” is not provided in the context, nical documents. Both AE and AD models could be used therefore, an AD system is needed to find the correct in downstream applications including information ex- meaning. Note that an acronym might refer to multiple The second workshop on Scientific Document Understanding at AAAI meanings. For instance in the above mentioned exam- 2022 ple, the acrony “CNN ” can be expanded to “Cable News Envelope-Open apouranb@cs.uoregon.edu (A. P. B. Veyseh) Network” or “Convolution Neural Network”. To correctly © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). select the right meaning for an ambiguous acronym, an CEUR Workshop Proceedings http://ceur-ws.org ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org) AD system should employ the context of the acronym the sentence and identify and find the correct meaning and other information regarding different meanings of of the ambiguous acronyms [28]. an acronym. Despite all progress so far on AD and AE, the majority Due to the importance of both AD and AE, in the liter- of the prior works are trained and evaluated on limited ature, there are various models proposed for each task. domains and languages. In particular, English and Bio- However, one limitation in the existing methods is that medicine are the predominant language and domain for they are trained and evaluated on specific languages and these tasks. This is a shortcoming as the challenges for domains. In particular, the majority of the existing AD AD and AE in other domains and languages are not ade- and AE resources are limited to English and biomedical or quately studied. To address this limitation, in this work, general domain. As such, the. challenges of these tasks in we propose a large scale acronym extraction and disam- other languages and domains are not adequately studied. biguation dataset in multiple languages and domains. To fill this gap, we present novel acronym extraction and disambiguation datasets that covers multiple languages and domains. In particular, for acronym extraction, we 3. Acronym Extraction collect and manually annotate documents in scientific We collect information in two spaces of legitimate and and legal domain in languages: English, Spanish, French, logical records for AE explanation. For each space, Danish, Persian and Vietnamese. For acronym disam- archives totally different dialects are required. As such, biguation task, we collect and automatically annotate for the legitimate space, we utilize the Joined together documents in scientific and legal domains in languages: Countries Parallel Corpus (UNPC) [29] and the Europarl English, Spanish and French. We also conduct two shared corpus [30]. The UNPC corpus contains official records tasks on the proposed dataset. In Acronym Extraction in 6 dialects whereas the Europarl corpus comprises of shared task, 58 teams participates and in Acronym Disam- the procedures of the European Parliament in European biguation shared task 44 teams participates. This paper dialects. To suit our comment budget and differentiate present the details of the dataset and the overview of the the coming about dataset, we select reports from four submitted systems for each task. dialects within the two corpora (i.e., English, French, and Spanish in UNPC, and Danish in Europarl) for our AE 2. Related work explanation. In expansion, for the scientific domain, we utilize the freely accessible papers and M.S./Ph.D. theses Acronym Extraction and Disambiguation are well known within the field of computer science for AE explanation. tasks for document understanding. In the last two Particularly, we collect the papers distributed within the decades, several methods have been proposed for AE or ACL collection of common dialect handling inquire about AD [11, 12, 13, 14, 15, 16, 17, 18]. Early works employed for English. Also, for typologically different languages, rule-based models. More specifically, a set of linguis- we crawl public computer science thesis in Persian and tic rules are defined to identify the acronyms and their Vietnamese. long-forms in text. Schwartz and Hearst [13] proposed To annotate the data, we hire freelancers from Upwork. to identify the long-forms and their acronyms based on The workers are fluent in the target language and have character match. That is, an acronym is labeled as the experience in data annotation. For a sentence in a dialect, short-form of a phrase if there are a sequence of char- we as it were comment on long shapes that are within the acters in the phrase that can form the acronym. Veyseh same dialect as the sentence’s. A short time later, for each et al. [19] extended the Schwartz’s rules by by identi- dialect, we hold two candidates who pass and accomplish fying the acronyms that are not accompanied by their most elevated comes about in our planned test for AE as long form. Later, feature engineering methods and deep our official annotators. Following, the two annotators in learning have been also employed for acronym extraction each dialect autonomously perform AE explanation for [20, 21]. Acronym disambiguation have been also exten- the inspected sentences of that dialect. At long last, the sively studied in the literature. This task can be modeled two annotators will examine to resolve any difference as a supervised classification task [22, 23, 24, 25, 26, 27]. within the comment, hence creating a last adaptation of Also, Zero-shot models, in which the long-forms of the our MACRONYM dataset [31]. The dataset statistics and acronym in test set are not seen by the models, have been agreement scores are presented in Table 1. proposed [19]. We conduct a shared task on Acronym Extraction at Moreover, in addition to the shared tasks presented SDU@AAAI-22 workshop. In this shared task, 58 teams in this work, SDU@AAAI-21 also hosted two shared participated in the task. Among which, 9 teams submit tasks on acronym identification and disambiguation. In their systems in the test phase. Table 2 shows the per- these shared tasks, the winning solutions employed deep formance of the participating systems in the test phase. learning models based on BERT transformer to encode Among all participating teams, “WENGSYX ” achieve the Domain IAA Size # Unique # Unique Team Language-Domain P R F1 & Language Acronyms Long-forms English-Legal 0.87 0.90 0.88 English 0.824 4,000 3,688 3,037 Spanish-Legal 0.90 0.91 0.91 Spanish 0.810 6,400 4,059 4,437 Legal French-Legal 0.93 0.92 0.92 French 0.823 8,000 5,638 5,728 WENGSYX Danish-Legal 0.95 0.98 0.96 Danish 0.810 3,000 907 923 Persian-Scientific 0.76 0.82 0.79 English 0.811 4,000 3,604 4,260 [32] Vietnamese-Scientific 0.85 0.82 0.84 Scientific Persian 0.782 1,000 641 203 [33] English-Scientific 0.85 0.87 0.86 Vietnamese 0.791 800 270 61 English-Legal 0.84 0.89 0.87 Spanish-Legal 0.90 0.91 0.90 Table 1 French-Legal 0.81 0.80 0.81 fazlfrs Danish-Legal 0.78 0.84 0.81 Statistics of Acronym Extraction dataset. IAA scores use Krip- Persian-Scientific 0.92 0.43 0.59 pendorff’s alpha with MASI distance based on initial inde- [35] Vietnamese-Scientific 0.37 0.36 0.36 pendent annotations. Size refers to the number of annotated English-Scientific 0.80 0.86 0.83 sentences. English-Legal 0.88 0.91 0.90 Spanish-Legal 0.90 0.90 0.90 French-Legal 0.92 0.93 0.93 LiSiheng Danish-Legal 0.95 0.95 0.95 highest score on four language-domain pairs (Spanish Persian-Scientific 0.69 0.53 0.60 [36] Vietnamese-Scientific 0.96 0.62 0.76 and Danish in legal and Persian and Vietnamese in sci- English-Scientific 0.89 0.89 0.89 entific domains). This model [32, 33] employs an adver- English-Legal 0.87 0.91 0.89 sarial training strategy. In particular, two methods are Spanish-Legal 0.90 0.90 0.90 employed for extracting the acronym and long-forms: (1) French-Legal 0.94 0.95 0.94 nithishkannen Danish-Legal 0.95 0.97 0.96 Sequence labeling, the task is modeled as sequence classi- [34] Vietnamese-Scientific 0.83 0.84 0.83 fication in BIO format. To this end, a BILSTM+CRF model English-Scientific 0.83 0.88 0.86 is employed. (2) Spand Detection: In this method the English-Legal 0.78 0.81 0.80 Spanish-Legal 0.87 0.90 0.88 acronyms and long-forms spans are directly predicted by French-Legal 0.77 0.76 0.77 the transformer-based model. “shihanmax” achieve best uyaseen Danish-Legal 0.89 0.90 0.89 performance on English test set for both scientific and le- Persian-Scientific 0.58 0.54 0.56 [37] Vietnamese-Scientific 0.48 0.67 0.56 gal domain, and “nithishkannen” has the highest score on English-Scientific 0.75 0.74 0.74 French legal domain. This model [34] employs character- English-Legal 0.75 0.69 0.72 level BERT model to address the out-of-vocabulary issues Spanish-Legal 0.65 0.65 0.65 which is restricting for acronym extraction. French-Legal 0.68 0.59 0.63 dipteshkanojia Danish-Legal 0.78 0.70 0.74 From Table 2, is is evident that the performance of the Persian-Scientific 0.64 0.51 0.57 models in scientific domain is lower than their perfor- [38] Vietnamese-Scientific 0.64 0.66 0.65 mance on legal domain. This performance drop indicates English-Scientific 0.77 0.69 0.73 English-Legal 0.90 0.42 0.57 the challenges in the scientific domain. Also, the lower Spanish-Legal 0.92 0.49 0.64 performance of the models in non-English languages, guneetsk99 French-Legal 0.89 0.35 0.50 specifically Persian and Vietnamese, reveal the challeng- Danish-Legal 0.90 0.45 0.60 English-Scientific 0.90 0.48 0.62 ing nature of AE in non-English languages. TC_AI_Lab Danish-Legal 0.09 0.06 0.07 English-Legal 0.90 0.92 0.91 shihanmax English-Scientific 0.89 0.92 0.90 4. Acronym Disambiguation Table 2 In addition to AE, an acronym understanding system Performance of the participating teams in test phase of should be able to find the correct meaning of the acronym extraction task, in terms of precision, recall and F1. The highest F1 score for each Language-Domain is in acronyms that are not accompanied with their long-form. bold-face. To evaluate the performance of the systems for this task, we automatically construct a dataset for acronym disam- biguation task. More specifically, given the annotations In this model [39, 40] a multi-choice approach is em- for the AE dataset, for every acronym in a document ployed for acronym disambiguation. In particular, the that is expanded to a long-form, we employed its pro- input sentence containing the ambiguous acronym along vided long-form as the label for any other mention of the with all possible expansions are provided to the model acronym in the given document (i.e., one meaning per dis- via different channels. Each expansion is scorees sep- course assumption). Using this approach, we construct a arately. Finally a unified model is employed to select dataset on English (legal and scientific domain), French - the expansion with the highest score. From Table 4, it is Legal and Spanish - Legal. The statistics of the dataset evident that models obtain higher score on English Sci- are presented in Table 4. In this shared task, “WENGSYX ” entific compared to other splits (i.e., legal test sets). This achieve the highest score on all languages and domains. Team Language-Domain P R F1 the test phase. English-Legal 0.94 0.87 0.90 French-Legal 0.89 0.79 0.84 WENGSYX Spanish-Legal 0.91 0.85 0.88 [39, 40] English-Scientific 0.97 0.94 0.96 5. Conclusion English-Legal 0.82 0.80 0.81 French-Legal 0.85 0.73 0.78 In this work, we presented two new acronym under- csyantins Spanish-Legal 0.88 0.79 0.83 standing resources in multiple languages and domains. English-Scientific 0.95 0.90 0.93 In particular, we presented manually annotated acronym English-Legal 0.86 0.77 0.81 French-Legal 0.81 0.72 0.76 extraction dataset in two domains of scientific and le- ghsong Spanish-Legal 0.86 0.77 0.81 gal documents and in six languages of English, Spanish, [41] English-Scientific 0.88 0.82 0.85 French, Danish, Persian, and Vietnamese. Moreover, we English-Legal 0.79 0.64 0.70 presented a novel automatically annotated dataset for French-Legal 0.76 0.70 0.73 TianHongZXY Spanish-Legal 0.83 0.80 0.81 acronym disambiguation in scientific and legal domain [42] English-Scientific 0.81 0.77 0.79 and in English, Spanish, and French. Using the proposed English-Legal 0.78 0.57 0.66 dataset, we conduct two shared tasks on acronym ex- French-Legal 0.73 0.64 0.68 traction and disambiguation. For each task, 9 and 11 TTaki Spanish-Legal 0.76 0.66 0.70 English-Scientific 0.81 0.69 0.75 teams participates in different domains and language. English-Legal 0.75 0.61 0.67 The performance of the winning systems, especially in mozhiwen French-Legal 0.72 0.63 0.67 non-English languages and legal domain, indicates the Spanish-Legal 0.86 0.80 0.83 necessity of further research on this task. English-Scientific 0.79 0.69 0.74 Decalogue English-Scientific 0.71 0.60 0.65 [43] References sherlock314159 English-Legal 0.70 0.59 0.64 [1] Y. Liu, F. Meng, J. Zhang, J. Xu, Y. Chen, J. Zhou, [44] Gcdt: A global context enhanced deep transition English-Legal 0.74 0.45 0.56 architecture for sequence labeling, in: ACL, 2019. kumudlakara Spanish-Legal 0.62 0.35 0.45 English-Legal 0.73 0.56 0.63 [2] A. Pouran Ben Veyseh, T. H. Nguyen, D. Dou, Graph English-Legal 0.78 0.47 0.58 based neural networks for event factuality predic- AbhayShukla French-Legal 0.70 0.49 0.57 tion using syntactic and semantic structures, in: Spanish-Legal 0.77 0.54 0.64 ACL, 2019. English-Scientific 0.73 0.52 0.61 English-Legal 0.98 0.04 0.07 [3] C. F. Ackermann, C. E. Beller, S. A. Boxwell, E. G. kyspid Katz, K. M. Summers, Resolution of acronyms English-Scientific 0.86 0.26 0.40 in question answering systems, 2020. US Patent Table 3 Performance of the participating teams in test phase of the 10,572,597. acronym disambigaution task, in terms of precision, recall [4] A. P. B. Veyseh, Cross-lingual question answering and F1. The highest F1 score for each language-domain is in using common semantic space, in: Proceedings bold-face. of TextGraphs-10: the workshop on graph-based methods for natural language processing, 2016, pp. Language & Domain Train Dev Test 15–19. English - Legal 2949 385 383 [5] A. P. B. Veyseh, F. Dernoncourt, D. Dou, T. H. English - Scientific 7532 894 574 Nguyen, A joint model for definition extraction French - Legal 7851 909 813 with syntactic connection and semantic consis- Spanish - Legal 6267 818 862 tency., in: AAAI, 2020, pp. 9098–9105. Table 4 [6] S. Spala, N. Miller, F. Dernoncourt, C. Dockhorn, Statistics of acronym disambiguation dataset SemEval-2020 task 6: Definition extraction from free text with the DEFT corpus, in: Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2020. higher performance indicates that in scientific domain, [7] S. Spala, N. A. Miller, Y. Yang, F. Dernoncourt, the acronyms are less ambiguous than the legal domain. C. Dockhorn, DEFT: A corpus for definition extrac- Using the prepared dataset, we conduct a shared task tion in free- and semi-structured text, in: Proceed- on acronym disambiguation at SDU@AAAI-22 workshop. ings of the 13th Linguistic Annotation Workshop, In this shared task, 44 teams participated. Among which, 2019. 11 teams submitted their system in the test phase. Table [8] L. Espinosa-Anke, S. Schockaert, Syntactically 3 shows the performance of the participating teams in aware neural architectures for definition extraction, [22] Y. Wang, K. Zheng, H. Xu, Q. Mei, Clinical word in: NAACL-HLT, 2018. sense disambiguation with interactive search and [9] Y. Jin, M.-Y. Kan, J.-P. Ng, X. He, Mining scientific classification, in: AMIA Annual Symposium Pro- terms and their definitions: A study of the ACL ceedings, volume 2016, American Medical Informat- Anthology, in: EMNLP, 2013. ics Association, 2016, p. 2062. [10] V. D. Lai, A. P. B. Veyseh, F. Dernoncourt, T. H. [23] Y. Li, B. Zhao, A. Fuxman, F. Tao, Guess me if you Nguyen, SemEval-2022 task 13: Symlink: Linking can: Acronym disambiguation for enterprises, in: mathematical symbols to their descriptions, in: Pro- Proceedings of the 56th Annual Meeting of the As- ceedings of the Fourteenth Workshop on Semantic sociation for Computational Linguistics (Volume 1: Evaluation, 2022. Long Papers), Association for Computational Lin- [11] Y. Park, R. J. Byrd, Hybrid text mining for finding guistics, Melbourne, Australia, 2018, pp. 1308–1317. abbreviations and their definitions, in: Proceedings URL: https://www.aclweb.org/anthology/P18-1121. of the 2001 conference on empirical methods in doi:1 0 . 1 8 6 5 3 / v 1 / P 1 8 - 1 1 2 1 . natural language processing, 2001. [24] Y. Wu, J. Xu, Y. Zhang, H. Xu, Clinical abbreviation [12] J. D. Wren, H. R. Garner, Heuristics for identifi- disambiguation using neural word embeddings, in: cation of acronym-definition patterns within text: Proceedings of BioNLP 15, 2015, pp. 171–176. towards an automated construction of comprehen- [25] R. Antunes, S. Matos, Biomedical word sense disam- sive acronym-definition dictionaries, Methods of biguation with word embeddings, in: International information in medicine 41 (2002) 426–434. Conference on Practical Applications of Computa- [13] A. S. Schwartz, M. A. Hearst, A simple algorithm for tional Biology & Bioinformatics, Springer, 2017, pp. identifying abbreviation definitions in biomedical 273–279. text, in: Biocomputing 2003, World Scientific, 2002, [26] J. Charbonnier, C. Wartena, Using word embed- pp. 451–462. dings for unsupervised acronym disambiguation, in: [14] E. Adar, Sarad: A simple and robust abbreviation Proceedings of the 27th International Conference dictionary, Bioinformatics 20 (2004) 527–533. on Computational Linguistics, Association for Com- [15] D. Nadeau, P. D. Turney, A supervised learning putational Linguistics, Santa Fe, New Mexico, USA, approach to acronym identification, in: Conference 2018. URL: https://www.aclweb.org/anthology/C18- of the Canadian Society for Computational Studies 1221. of Intelligence, Springer, 2005, pp. 319–329. [27] M. R. Ciosici, T. Sommer, I. Assent, Unsuper- [16] H. Ao, T. Takagi, Alice: an algorithm to extract ab- vised abbreviation disambiguation, arXiv preprint breviations from medline, Journal of the American arXiv:1904.00929 (2019). Medical Informatics Association 12 (2005) 576–586. [28] A. P. B. Veyseh, F. Dernoncourt, T. H. Nguyen, [17] K. Kirchhoff, A. M. Turner, Unsupervised resolution W. Chang, L. A. Celi, Acronym identification and of acronyms and abbreviations in nursing notes us- disambiguation shared tasks for scientific document ing document-level context models, in: Proceedings understanding, arXiv preprint arXiv:2012.11760 of the Seventh International Workshop on Health (2020). Text Mining and Information Analysis, 2016, pp. [29] M. Ziemski, M. Junczys-Dowmunt, B. Pouliquen, 52–60. The united nations parallel corpus v1.0, in: [18] A. P. B. Veyseh, F. Dernoncourt, Q. H. Tran, T. H. Proceedings of the Tenth International Confer- Nguyen, What does this acronym mean? introduc- ence on Language Resources and Evaluation ing a new dataset for acronym identification and LREC 2016, Portorož, Slovenia, May 23-28, 2016, disambiguation, arXiv preprint arXiv:2010.14678 2016. URL: http://www.lrec-conf.org/proceedings/ (2020). lrec2016/summaries/1195.html. [19] A. P. B. Veyseh, F. Dernoncourt, W. Chang, T. H. [30] P. Koehn, Europarl: A parallel corpus for Nguyen, Maddog: A web-based system for acronym statistical machine translation, in: Proceed- identification and disambiguation, arXiv preprint ings of Machine Translation Summit X: Papers, arXiv:2101.09893 (2021). MTSummit 2005, Phuket, Thailand, September [20] C.-J. Kuo, M. H. Ling, K.-T. Lin, C.-N. Hsu, Bioadi: a 13-15, 2005, 2005. URL: https://aclanthology.org/ machine learning approach to identifying abbrevia- 2005.mtsummit-papers.11. tions and definitions in biological literature, in: [31] A. P. B. Veyseh, N. Meister, S. Yoon, R. Jain, F. Der- BMC bioinformatics, volume 10, Springer, 2009, noncourt, T. H. Nguyen, Macronym: A large-scale p. S7. dataset for multilingual and multi-domain acronym [21] J. Liu, C. Liu, Y. Huang, Multi-granularity sequence extraction, arXiv preprint arXiv:2202.09694 (2022). labeling model for acronym expansion identifica- [32] X. Huang, B. Li, F. Xia, Y. Weng, A novel initial tion, Information Sciences 378 (2017) 462–474. reminder framework for acronym extraction, in: SDU@AAAI-22, 2022. [33] B. Li1, F. Xia, Y. Weng, X. Huang, B. Sun, S. Li, Psg: Prompt-based sequence generation for acronym extraction, in: SDU@AAAI-22, 2022. [34] N. Kannen, D. Sheth, A. Chandra, S. Pal, Cabace: In- jecting character sequence information and domain knowledge for enhanced acronym and long-form extraction, in: SDU@AAAI-22, 2022. [35] Balouchzahi, Vitman, Shashirekha, Sidorov, Gel- bukh, Acronym identification using transformers and flair framework, in: SDU@AAAI-22, 2022. [36] S. Li, C. Yang, T. Liang, X. Zhu, C. Yu, Y. Yang, Acronym extraction with hybrid strategies, in: SDU@AAAI-22, 2022. [37] S. L. Usama Yaseen, Domain adaptive pretrain- ing for multilingual acronym extraction, in: SDU@AAAI-22, 2022. [38] P. Sharma, H. Saadany, L. Zilio, D. Kanojia, C. Orasan, An ensemble approach to acronym ex- traction using transformers, in: SDU@AAAI-22, 2022. [39] Y. Weng, F. Xia, B. Li, X. Huang, S. He, Adbcmm : Acronym disambiguation by building counterfac- tuals and multilingual mixing, in: SDU@AAAI-22, 2022. [40] B. Li, F. Xia, Y. Weng, X. Huang, B. Sun, Simclad: A simple framework for contrastive learning of acronym disambiguation, in: SDU@AAAI-22, 2022. [41] G. Song, H. Lee, K. Shim, T5 encoder based acronym disambiguation with weak supervision, in: SDU@AAAI-22, 2022. [42] X. Zhu, C. Yu, S. Li, T. Liang, C. Yang, Y. Yang, Multilingual acronym disambiguation with multi- choice classification, in: SDU@AAAI-22, 2022. [43] Y. Y. Taiqiang Wu, Xingyu Bai, Prompt-based model for acronym disambiguation via negative sampling, in: SDU@AAAI-22, 2022. [44] F. Xia, B. Li, X. Huang, Y. Weng, S. He, Anaconda: Adversarial training with in-trust loss in acronym disambiguation, in: SDU@AAAI-22, 2022.