<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gautam-2026-lipi">
    <titleInfo>
        <title>LIPI: Transliterating Hindi the Way People Type It</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Abhishek</namePart>
        <namePart type="family">Gautam</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Toshiaki</namePart>
            <namePart type="family">Nakazawa</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Isao</namePart>
            <namePart type="family">Goto</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hengqin, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-446-0</identifier>
    </relatedItem>
    <abstract>Romanized Hindi has no standard spelling: the same word may be typed nahi, nahin, or nai, and no formal romanization rule or general-purpose transliteration system captures this typed convention well. Transliteration quality therefore depends as much on which spelling a user would actually type as on phonological plausibility. We present LIPI, a compact, encoder-only Devanagari-to-Roman transliteration model (~0.5M parameters) that represents each akshara compositionally from its ordered Unicode codepoints, expands it into six learned CTC emission slots, and encodes the frames with a two-layer bidirectional GRU under a connectionist-temporal-classification objective. The model is trained on a usage-ranked lexicon we build only from the training partitions of Dakshina, Xlit-Crowd, and Aksharantar, ranked by an attestation-gated social-usage prior so that a frequent Roman spelling is preferred only when it is also attested as a transliteration of the source word. LIPI is on par with the much larger multilingual IndicXlit baseline on general word transliteration (Aksharantar) and far stronger on naturally typed queries (FIRE 2013: 73.2 vs. 34.8 exact match), roughly doubling both baselines’ word match on Dakshina and IndoNLP sentences while remaining small enough for on-device use. The word-level gains are statistically significant and hold on words unseen in training, and an emission-slot ablation isolates the learned expansion’s contribution. The model, lexicon, and full benchmark results are publicly released at https://github.com/ABHISHEKgauti25/lipi.</abstract>
    <identifier type="citekey">gautam-2026-lipi</identifier>
    <location>
        <url>https://aclanthology.org/2026.wat-1.2/</url>
    </location>
    <part>
        <date>2026-11</date>
        <extent unit="page">
            <start>9</start>
            <end>16</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T LIPI: Transliterating Hindi the Way People Type It
%A Gautam, Abhishek
%Y Nakazawa, Toshiaki
%Y Goto, Isao
%S Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-446-0
%F gautam-2026-lipi
%X Romanized Hindi has no standard spelling: the same word may be typed nahi, nahin, or nai, and no formal romanization rule or general-purpose transliteration system captures this typed convention well. Transliteration quality therefore depends as much on which spelling a user would actually type as on phonological plausibility. We present LIPI, a compact, encoder-only Devanagari-to-Roman transliteration model (~0.5M parameters) that represents each akshara compositionally from its ordered Unicode codepoints, expands it into six learned CTC emission slots, and encodes the frames with a two-layer bidirectional GRU under a connectionist-temporal-classification objective. The model is trained on a usage-ranked lexicon we build only from the training partitions of Dakshina, Xlit-Crowd, and Aksharantar, ranked by an attestation-gated social-usage prior so that a frequent Roman spelling is preferred only when it is also attested as a transliteration of the source word. LIPI is on par with the much larger multilingual IndicXlit baseline on general word transliteration (Aksharantar) and far stronger on naturally typed queries (FIRE 2013: 73.2 vs. 34.8 exact match), roughly doubling both baselines’ word match on Dakshina and IndoNLP sentences while remaining small enough for on-device use. The word-level gains are statistically significant and hold on words unseen in training, and an emission-slot ablation isolates the learned expansion’s contribution. The model, lexicon, and full benchmark results are publicly released at https://github.com/ABHISHEKgauti25/lipi.
%U https://aclanthology.org/2026.wat-1.2/
%P 9-16

Markdown (Informal)

[LIPI: Transliterating Hindi the Way People Type It](https://aclanthology.org/2026.wat-1.2/) (Gautam, WAT 2026)

• LIPI: Transliterating Hindi the Way People Type It (Gautam, WAT 2026)

ACL

• Abhishek Gautam. 2026. LIPI: Transliterating Hindi the Way People Type It. In Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026), pages 9–16, Hengqin, China. Association for Computational Linguistics.