{ "fetched_at": "2026-10-19T19:05:46Z", "source_url": "https://www.kaggle.com/competitions/dataverse_2023/data", "pages": [ { "name": "data-description", "content": "In the Bengali language, there are seven vowel phonemes: ɐ, ɛ, e, i, o, ɔ, u. For example, আঘাত which means 'injury’ is transcribed as /ɐgʱɐt̪/, আশা as /ɐʃɐ/, অ্যাকাউন্ট as /ɛkɐu̯nt/. We have employed the phonetic chart created by Peter Ladefoged[1] to represent the phonetic sounds of the Bengali language.\n\nThe Bengali consonant phonemes are: k, kʰ, g, gʰ, ŋ, c, cʰ, ɟ, ɟʰ, t, tʰ, d, dʱ,n, t̪, t̪ʰ, d̪, d̪ʱ, p, pʰ, b, bʱ, m, ɾ, l, ʃ, s, h, ɽ, ɽʰ. While some Bengali linguists suggested using affricate sounds for চ/tʃ/, ছ/tʃʰ/, জ/dʒ/, ঝ/dʒʱ/, we're recommending plosive sounds for our dataset which is more appropriate in the context of Bengali language. For the ট and ঠ consonant, we are using the alveolar /t/, /tʰ/ sounds. For example, টমেটো /tɔmeto/, চেতনা /cet̪onɐ/ ঘড়ি /gʱoɽɪ/, জরিপে /ɟoɾɪpe/ and so on. \n\nDue to having many borrowed words from languages like Arabic, Chinese, English, Farsi, Portuguese, etc. in the Bengali language, the speaker affects the pronunciation of these borrowed words. In standard Bengali, people tend to use the 'z' sound instead of 'j' and the 'f' sound instead of 'pʰ' which is not common in Bengali consonant sounds. For instance, \"fɐt̪emɐ\" is pronounced instead of \"pʰɐt̪emɐ\", and \"fɐɾmesɪ\" is used instead of \"pʰɐɾmesɪ\". Similarly, \"zɐɪd̪\" is used instead of \"ɟɐɪd̪\", and so forth.\n\nBelow are some IPA transcribed Bengali sentences:\n১৯ নভেম্বর নরেন্দ্র মোদি স্টেডিয়াম, বাংলাদেশ বিশ্বকাপ শিরোপা তুললো /unɪʃ novembɔɾ nɔɾendɾɐ modɪ stedɪum, bɐŋlɐd̪eʃ bɪʃʃokɐp ʃɪɾopɐ t̪ullo/ \nএক মাঘে শীত নাহি যায় \nকথায় বলে কথায় থাক /kɔt̪ʰae̯ bɔle kɔt̪ʰae̯ tʰɐk/\nকমলা সুন্দরীর পিন্দনের শাড়ি \nসৎকর্ম বৃথা যায় না /ʃɔt̪kɔrmo bɾɪt̪ʰa ɟae̯ na/ \n\nWe have proposed the /ɐ/ phoneme instead of /a/ and /ɛ/ instead of /æ/ which are more commonly used to transcribe Bengali language.\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "## Goal of the Competition\nThe goal of this competition is to recognize model IPA transcription from Bengali texts(Remember the greek characters in the dictionary, to help you find out the accurate pronunciation of words? That was International Phonetic Alphabet (IPA) transcription! You will build a model trained on a linguist validated dataset containing Bengali text from different domains. The test set contains numbers, loan-words and domain-specific words to add to the challenge.\n\nYour efforts could improve Bengali computational linguistics and NLP research using the first Bengali sentence level IPA transcription dataset from Bengali.AI . In addition, your submission will be among the first open-source IPA transcription methods for Bengali.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "Submissions are evaluated by a mean [**Word Error Rate**](https://en.wikipedia.org/wiki/Word_error_rate), proceeding as follows:\n1. The WER is computed for each instance in the test set.\n2. The WERs are averaged within domains, weighted by the number of words in the sentence.\n3. The (unweighted) mean of the domain averages is the final score.\n\nThis Python code computes the metric:\n```python\nimport jiwer # you may need to install this library\n\ndef mean_wer(solution, submission):\n joined = solution.merge(submission.rename(columns={'sentence': 'predicted'}))\n domain_scores = joined.groupby('domain').apply(\n # note that jiwer.wer computes a weighted average wer by default when given lists of strings\n lambda df: jiwer.wer(df['sentence'].to_list(), df['predicted'].to_list()),\n )\n return domain_scores.mean()\n```\n## Submission Format\n\nThe submission files should contain two columns: `row_id_column_name` and `ipa`. You will need to predict the sentence for each sentence in the test dataset file. \n\nThe submission file should contain a header and have the following format:\n\nrow_id_column_name,ipa\n41,unɪʃ novembɔɾ nɔɾendɾɐ modɪ stedɪum, bɐŋlɐd̪eʃ bɪʃʃokɐp ʃɪɾopɐ t̪ullo\n42,unɪʃ novembɔɾ nɔɾendɾɐ modɪ stedɪum, bɐŋlɐd̪eʃ bɪʃʃokɐp ʃɪɾopɐ t̪ullo\n43,unɪʃ novembɔɾ nɔɾendɾɐ modɪ stedɪum, bɐŋlɐd̪eʃ bɪʃʃokɐp ʃɪɾopɐ t̪ullo", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "# Registration \n**Link:** https://myeventspark.com/events/view/dataverse-challenge-(datathon).\n**Deadline:** October 15, 2023\n# Team Requirements\n- A team must consist of at most 3 members. \n- The members of a team can be from different universities. \n- Each team must have at least one undergraduate member, and they may also include up to one graduate participant (industry professional/senior member) if desired. \n- For overseas participants, a minimum of 1 person should be present in Bangladesh during the final presentation.\n\n# Rules for the Competition\n- Registration is mandatory to participate in all the phases of the contest.\n- Only one account per participant in Kaggle will be allowed.\n- Participants are allowed to use external data, but the data has to be disclosed publicly before the final round. \n- The use of the internet is allowed for research and reference, but API calls are not permitted.\n- All teams must submit working Kaggle notebooks for training and inference (and the models too), including appropriate instructions and documentation for reproducibility. The inference notebook must run on Kaggle Notebooks. \n- Teams are required to make all scripts, data, and reports publicly available.\n- No private code sharing outside teams.\n- Teams must prepare a project report of at least 2 pages. The paper has to be in IEEE/ACM (2 column) format.\n\n## General Rules\n\n- The DataVerse Challenge Organizing Committee retains the authority to cancel or reschedule the event under any circumstances. The organizers serve the right to disqualify any team.\n- The selection of winners will be made solely at the discretion of the DataVerse Challenge Organizing Committee. All participants are required to adhere to the terms and conditions. Decisions made by the DataVerse Challenge Organizing Committee in this matter will be conclusive, and no requests for clarification or appeals will be considered.\n- The participants must only use open source code licensed under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code.\n- Winner license: The winners will grant the competition hosts and sponsors the following license: \nYou hereby license and will license your winning submission and the source code used to generate the submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate the submission that is not owned by you but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software. You must also make your solution public.\nThe Competition hosts and sponsors and its affiliates may use your name and likeness for advertising and promotional purposes without additional compensation unless prohibited by law.\n- You warrant that your submission is your own original work and, as such, you are the sole and exclusive owner and rights holder of the submission, and you have the right to make the submission and grant all required licenses. To the maximum extent permitted by law, you indemnify and agree to keep indemnified Competition Entities at all times from and against any liability, claims, demands, losses, damages, costs, and expenses resulting from any of your acts, defaults or omissions and/or a breach of any warranty set forth herein.\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "abstract", "content": "**Bengali Text to IPA (International Phonetic Alphabet) Transcription** is an area that has seen relatively limited development compared to other languages, despite Bengali being one of the world's most widely spoken native languages. There is a growing need for automated systems that can accurately convert Bengali text into IPA notation due to the vast audience and various applications in linguistics, language learning, and phonetic research. Having this in mind, we welcome participants to participate in **DataVerse**, a part of **ITVerse 2023** organized by **IIT Software Engineers' Community (IITSEC)** as we partner with **Bengali.AI** to advance research in Bengali text to IPA domain.\n\n## How to Participate\nRegistration to the **DataVerse** challenge is completely free. Anyone can join the competition. However, registration is mandatory to be eligible for the final round.\n***Please refer to the [Rules section](https://www.kaggle.com/competitions/dataverse_2023/rules) For team formation guidelines***\nRegistration link: https://myeventspark.com/events/view/dataverse-challenge-(datathon)\n\nFacebook event link: https://fb.me/e/4tQ5nxYmu\n\nJoin Facebook group to connect with fellow participants and the organizing committee: https://www.facebook.com/groups/1410611209517321/\n\n## Competition Schedule\n**Initial Round:** Will be held on Kaggle from October 9th to November 1st. Participants are expected to build and train models during this phase. The top 15 registered teams from this round will be invited to join us in the final round.\n**Final Round:** Will be held onsite at the Institute of Information Technology, University of Dhaka on 5th November. Only invited teams will get a chance to present their work in front of the judge's panel, where they will have to submit a IEEE/ACM (2 column) paper. Furthermore, their inference notebooks will be evaluated on a hidden dataset.\n\n\n## Scoring\nThe scoring weights are divided in the following way - \n**Round 1 Public Standing** : 12%\n**Round 1 Private Standing** : 18%\n**Hidden Evaluation** : 50%\n**Paper & Presentation** : 20%\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }