[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85188-en":3,"doc-seo-85188-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85188,962075114101,"Seraphina","https://ap-avatar.wpscdn.com/avatar/e000253a75eb197efd?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780044092746381165",8,"Research & Report","Which Languages Transfer Best to Warlpiri Similarity-Based Study for Low-Resource ASR","This paper investigates how language similarity improves cross-lingual transfer for automatic speech recognition (ASR) in extremely low-resource settings. Warlpiri, an Australian Aboriginal language with very limited transcribed speech data, requires transfer learning to enable robust modeling. A framework ranks candidate high-resource source languages using acoustic similarity from pre-trained speech models and linguistic similarity from typology, phoneme inventories, grammatical, and syntactic features. Whisper-based experiments show acoustic and typological similarity outperform baselines, with Assamese and Hindi reducing word and character error rates.","Which Languages Transfer Best to Warlpiri? A Similarity-Based Study for  \nLow-Resource ASR  \nPravina Mylvaganam  1 , Eliathamby Ambikairajah 1, Ting Dang2 , Vidhyasaharan Sethu 1, Tuende  \nSzalay3  \n1 University of New South Wales, Australia  \n2 University of Melbourne, Australia  \n3 University of Sydney, Australia  \n[p.mylvaganam@unsw.edu.au](p.mylvaganam@unsw.edu.au) , [e.ambikairajah@unsw.edu.au](e.ambikairajah@unsw.edu.au) , [ting.dang@unimelb.edu.au](ting.dang@unimelb.edu.au) , [v.sethu@unsw.edu.au](v.sethu@unsw.edu.au) , [tuende.szalay@sydney.edu.au](tuende.szalay@sydney.edu.au)  \narXiv :2607 . 10256v 1 [ cs .CL] 11 Jul 2026  \nAbstract  \nThis paper investigates how language similarity can improve cross-lingual transfer for automatic speech recognition (ASR) in extremely low-resource settings. Warlpiri, an Australian Aboriginal language, has very limited transcribed speech data, making transfer learning essential. We propose a framework combining acoustic similarity from pre-trained speech models with linguistic similarity based on typology, phoneme inventories, grammatical, and syntactic features to rank highresource source languages and evaluate their effectiveness for ASR transfer to Warlpiri. Experiments with Whisper show that acoustically and typologically similar languages outperform monolingual and multilingual baselines. Assamese and Hindi achieve substantial reductions in word and character er  \nror rates. Correlation analysis further indicates that acoustic similarity is the strongest predictor of fine-tuning performance, while phoneme inventory and typological similarity better explain zero-shot transfer.  \nIndex Terms: low-resource ASR, cross-lingual transfer, language similarity, speech embeddings, Aboriginal languages  \n1. Introduction  \nAutomatic Speech Recognition (ASR) has become a transformative technology that enables natural human–computer interaction in applications such as virtual assistants, language learning, and accessibility tools. In recent years, ASR has advanced markedly due to large-scale annotated datasets and powerful deep learning models [1] . However, many languages, especially Indigenous and low-resource languages, remain underrepresented and perform worse than high-resource languages due to the limited transcribed corpora [2] . Warlpiri, an Australian Aboriginal language spoken by a small community in the Northern Territory [3], exemplifies this challenge, as its extremely limited data hinder robust ASR development.  \nCross-lingual transfer has emerged as a promising solution to data scarcity, leveraging acoustic and linguistic knowledge from high-resource languages to support low-resource ones [4, 5] . However, its effectiveness depends on the selection of suitable source languages for knowledge transfer [6] . Existing studies often choose source languages based on data availability, language families, or heuristics [7, 8, 9, 10], which may not reflect true acoustic or linguistic similarity [11, 12] . Recent work instead highlights acoustic and phonetic similarity as more reliable indicators [13], often yielding better transfer than genealogical or geographical closeness alone [14, 15], given that geographically or genealogically related languages can still differ substantially in their acoustic properties [16, 17, 18] . This  \nhighlights the need for systematic, similarity-based source language selection to maximize transfer in low-resource settings.  \nWarlpiri is typologically distant (i.e., differing substantially in grammar, morphology, and syntax) from most high-resource languages [19, 20, 21] and has received limited attention, making it unclear which high-resource languages are the most suitable transfer sources. Its segmental and suprasegmental properties further underscore the need for a systematic similarity analysis: the vowel system of Warlpiri is relatively small, while its consonant inventory is extensive, including dental, alveolar, retroflex, and palatal sounds that are","cbCaitlC1UkcUdgQ","https://ap.wps.com/l/cbCaitlC1UkcUdgQ","pdf",1211592,2,1,6,"English","en",105,"# Introduction\n## Motivation: ASR for low-resource and Indigenous languages\n## Problem: selecting effective transfer source languages\n## Focus: Warlpiri and similarity-based analysis\n## Related work and limitations","[{\"question\":\"Why is cross-lingual transfer important for Warlpiri ASR?\",\"answer\":\"Warlpiri has extremely limited transcribed speech data, which makes robust ASR development difficult. Cross-lingual transfer uses knowledge from high-resource languages to support low-resource modeling.\"},{\"question\":\"What two types of similarity does the study use to rank source languages?\",\"answer\":\"The study uses embedding-based acoustic similarity from pre-trained speech models and linguistic-feature-based similarity derived from typology, phoneme inventories, grammatical properties, and syntactic characteristics.\"},{\"question\":\"Which similarity factors most strongly predict fine-tuning and zero-shot transfer performance?\",\"answer\":\"Correlation analysis indicates acoustic similarity is the strongest predictor of fine-tuning performance. For zero-shot transfer, phoneme inventory and typological similarity better explain performance.\"}]",1784201632,15,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"which-languages-transfer-best-to-warlpiri-similarity-based-study-for-low-resource-asr","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/which-languages-transfer-best-to-warlpiri-similarity-based-study-for-low-resource-asr/85188/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-21","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why is cross-lingual transfer important for Warlpiri ASR?","Question",{"text":75,"@type":76},"Warlpiri has extremely limited transcribed speech data, which makes robust ASR development difficult. Cross-lingual transfer uses knowledge from high-resource languages to support low-resource modeling.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What two types of similarity does the study use to rank source languages?",{"text":80,"@type":76},"The study uses embedding-based acoustic similarity from pre-trained speech models and linguistic-feature-based similarity derived from typology, phoneme inventories, grammatical properties, and syntactic characteristics.",{"name":82,"@type":73,"acceptedAnswer":83},"Which similarity factors most strongly predict fine-tuning and zero-shot transfer performance?",{"text":84,"@type":76},"Correlation analysis indicates acoustic similarity is the strongest predictor of fine-tuning performance. For zero-shot transfer, phoneme inventory and typological similarity better explain performance.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,127,130,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]