A POS Tier Is the Key to Automated Annotation for Low-Resource Language Documentation: Neural Interlinear Glossing for Irabu, a Southern Ryukyuan Language

Michinori Shimoji (Kyushu University)

REVIEW 2 major objections 4 minor 31 references

A per-morpheme POS tier materially improves automatic grammatical glossing when POS input is gold-quality and training data are scarce, but predicted-POS errors cancel the gain in a fully automatic pipeline.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 05:19 UTC pith:SSQ5FORP

load-bearing objection Clean ablation of the POS tier, honestly reported null for the automatic pipeline, but the headline data-multiplier claim rests on gold POS and a 405-token test set — worth a serious referee, not a desk reject. the 2 major comments →

arxiv 2607.13372 v1 pith:SSQ5FORP submitted 2026-07-15 cs.CL

A POS Tier Is the Key to Automated Annotation for Low-Resource Language Documentation: Neural Interlinear Glossing for Irabu, a Southern Ryukyuan Language

Michinori Shimoji (Kyushu University) This is my paper

classification cs.CL

keywords low-resource language documentationinterlinear glossingPOS taggingquadrilinear annotationIrabu Ryukyuanneural sequence labelingerror propagationannotation budget

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks a practical question for endangered-language documentation: given a fixed, small annotation budget, does adding a per-morpheme part-of-speech (POS) tier to the standard three-line interlinear format make automated glossing better? Using a deliberately small neural pipeline trained on roughly one hour of fully annotated Irabu Ryukyuan discourse, it shows that gold-quality POS raises grammatical-gloss accuracy by 4.4 points on average, and the gain grows to 11.6 points when the training corpus is cut to a quarter. The gain is not yet realized automatically: when POS tags are predicted by the pipeline's own tagger, the few errors actively mislead the glossing model, so a wrong POS is worse than no POS. The paper frames the value as latent rather than lost, shows that it returns once the tagger clears roughly 90% accuracy, and recommends that documentation projects annotate quadrilinearly—text, POS, gloss, translation—so the tier is available for both assisted and future automatic workflows.

Core claim

The paper claims that an explicitly recorded per-morpheme POS tier—the missing fourth line in conventional trilinear interlinear glossing—is the key to making automatic glossing useful in low-resource documentation. Under gold POS, a BiLSTM-CRF glossing model improves grammatical-gloss accuracy by +4.35±0.73 points over a POS-ablated model, with the effect significant in all five seeds and concentrated on ambiguous case and focus clitics like Irabu ga, whose gloss is determined by its own POS and the category of the adjacent word. The gain grows as the training corpus shrinks, reaching +11.6 points at a quarter of the data, so a POS tier more than halves the amount of glossed data needed to

What carries the argument

The central object is the per-morpheme POS tier added to standard interlinear glossing, making the annotation quadrilinear (text + POS + gloss + translation). The argument is carried by a controlled ablation: a BiLSTM-CRF glossing model is trained with and without a POS embedding, holding all other factors fixed, under three conditions—gold POS, predicted POS, and no POS. The machinery includes the POS-to-gloss disambiguation rules the model learns (e.g., Irabu ga is genitive before a noun but nominative before a verb; as a focus particle it forces the focus gloss), and an error-propagation decomposition that attributes the lost gain to a train/test mismatch: the model, trained exclusively o

Load-bearing premise

The quantitative conclusions rest on a single corpus of 774 utterances (6,412 morpheme tokens) of one language, Irabu Ryukyuan, with a test set of only 405 grammatical-gloss tokens, and the learning-curve and data-multiplier analyses use gold POS at test time, so those results describe the assisted-annotation scenario rather than the fully automatic pipeline.

What would settle it

Run the same controlled ablation (gold vs. no POS) on another agglutinative language with a comparable one-hour gold corpus: if the gold-POS gain is not positive and significant across five seeds, or if the gain does not grow as training data shrink, the central claim is contradicted. A cheaper check is to measure gloss accuracy on the 34 mistagged tokens in this corpus under a model trained with POS dropout or predicted-POS augmentation; if the 'wrong POS worse than no POS' effect disappears, the paper's mechanistic explanation is confirmed.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

Adding a POS tier gives an immediate accuracy boost in computer-assisted annotation, where a human supplies POS before gloss review, with roughly 40% fewer gloss errors to fix at full data.
A POS tier more than halves the amount of glossed data needed to reach a given gloss accuracy in the low-data regime, so documentation projects can invest in rich annotation of less discourse rather than thin annotation of more.
The learning-curve interaction implies that annotation richness (POS) is a more cost-effective investment than annotation quantity alone for agglutinative languages with comparable morphological transparency.
The break-even analysis sets a concrete target: once the tagger exceeds roughly 90% accuracy, the POS gain starts to pay off automatically; at 92–96% it recovers +1.6 to +3.2 points.
The two-stage design implies that projects should front-load about 30 minutes of quadrilinear annotation, then use the pipeline to draft annotations for review, folding corrections back into training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Because the POS tier is a closed, cheap inventory, the benefit may be larger for typologically similar languages where case and focus morphology are ambiguous from surface form alone; replicating the ablation on another agglutinative language would directly test the generality.
The poisoning result suggests model-side fixes the paper lists as future work—training with predicted or noised POS, joint POS-gloss modeling, or soft POS inputs—and one can hypothesize that scheduled sampling would recover most of the +4.4 point ceiling even with the current tagger.
The learning-curve analysis used gold POS at test time, so the 'POS as a data multiplier' conclusion applies to the assisted scenario; extending it to the fully automatic pipeline would require retraining the tagger at each data budget, which is likely to reduce but not eliminate the advantage.
For a project that cannot reach ~90% POS accuracy, spending annotation effort on the POS tier may not pay off in automatic glossing, though it still helps the assisted workflow; this is a caveat beyond the paper's own hedge.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Clean ablation of the POS tier, honestly reported null for the automatic pipeline, but the headline data-multiplier claim rests on gold POS and a 405-token test set — worth a serious referee, not a desk reject.

read the letter

Two things to know. The core ablation — gold POS vs no POS under a fixed architecture — is clean and carefully analyzed, and the paper is refreshingly honest that the fully automatic pipeline gains nothing yet. The headline that a POS tier 'more than halves' the supervised data needed is only supported for the assisted scenario, because the learning curves in Table 5 use gold POS at test time. The paper says this in §3.7 and Limitations, but the abstract and conclusion still state the data-multiplier result without the caveat.

What is new: treating the presence of a POS tier as an annotation-design variable, with five seeds, exact McNemar tests, an exact error-propagation decomposition, and a break-even analysis that gives a concrete target for tagger accuracy. The per-label results showing the gain concentrated on case and focus clitics are convincing. The worked example of how the model disambiguates 'ga' is a nice transparency exhibit. The null pipeline result is reported rather than buried; that is to the paper's credit.

Soft spots, in proportion. Everything rests on one language and a test set of 405 grammatical-gloss tokens; the numbers should be read as one measurement, not constants. The data-equivalence values (×2.8, ×2.5, 'not matched even with the full corpus') are presented without confidence intervals, and the flagship comparison — POS-informed on 310 utterances vs ablated on all 620 — is 0.890 vs 0.888, which is inside the seed noise you would expect from the reported SDs. The 'wrong POS is worse than no POS' finding is based on 34 mistagged tokens; the direction is plausible and the decomposition is elegant, but the magnitude is fragile. The pipeline-mode learning curves, where the tagger is retrained at each budget, are never run; so the 'richness beats quantity' recommendation for the fully automatic setting is a projection, not a measurement. The claim that adding a POS tier is nearly free is asserted, not measured — no annotation time data are given.

Who benefits: field linguists planning documentation workflows and anyone working on automated glossing. The practical recommendation — write down the POS tier — is reasonable and may save real time in the assisted setup. This is a serious, well-scoped case study; I would send it out rather than desk-reject. Revisions I'd ask for: either add a second language or narrow the headline to the assisted case, add intervals to the data-equivalence numbers, and preferably measure tagger accuracy at the small training budgets so the multiplier claim has a pipeline analogue.

Referee Report

2 major / 4 minor

Summary. The paper presents a fully supervised BiLSTM–CRF pipeline for morpheme segmentation, POS tagging, and grammatical glossing of Irabu Ryukyuan, trained on 47 minutes/620 utterances of a 774-utterance corpus, and evaluates how annotation richness (whether a per-morpheme POS tier is provided) and quantity (6–47 min budgets) affect glossing accuracy. The central findings are: gold POS improves test grammatical-gloss accuracy by +4.35±0.73 points over an ablated no-POS model, significant in all 5 seeds; the gain is larger at smaller training budgets (+11.6 at 12 minutes); a fully automatic pipeline does not realize the gain because the tagger's 12% errors poison the glossing model (wrong POS worse than no POS), with break-even tagger accuracy near 88%. The authors recommend quadrilinear annotation and a two-stage assisted-annotation workflow, and they release the code, corpus, and per-sentence outputs.

Significance. If the results hold, this is one of the few controlled, budget-centred measurements of annotation design for low-resource documentary linguistics. The core ablation is well controlled: same architecture, data splits, and seeds; exact McNemar tests; seed-level consistency; and an exact per-token decomposition of the pipeline loss. The paper also gives a falsifiable target for tagger quality (break-even near 88% accuracy) and is unusually candid about limitations, releasing code, gold corpus, and complete test outputs. The main quantitative claims rest on a single 405-token test set from one language, and the data-multiplier claim is computed in the gold-POS assisted setting rather than the fully automatic pipeline; these are load-bearing caveats for the paper's central recommendation.

major comments (2)

[Abstract; §3.7/Table 5; §4] The headline data-multiplier claim — 'a POS tier more than halves the amount of glossed data needed' (Abstract; Contribution iv) — is measured only under gold POS at test time (§3.7 explicitly says the curves use gold POS and that a pipeline analogue is future work). The paper's motivating scenario is automatic annotation of new recordings, where POS is predicted by a tagger. §3.6 shows break-even at ~88% tagger accuracy and that the current 88.1% tagger delivers only +0.40±0.77 points; tagger accuracy will not be higher at the 6–23-minute budgets where Table 5 shows the largest gains. Since the Conclusion's 'richness beats quantity' and the Stage-1 '30 rich minutes' recommendation depend on this claim, the abstract/conclusion must restrict it to the assisted setting, or the paper must add pipeline-mode learning curves (retrained tagger per budget).
[§3.7, Table 5] The data-equivalence values (×2.8, ×2.5, 'not matched even with full corpus') are point estimates without confidence intervals or paired significance tests. The crossover anchoring the recommendation — Gold-POS at 310 utterances (0.890) vs No-POS at 620 (0.888) — is 0.2 points, within the seed SDs reported in Table 2. Report per-seed values and a paired/bootstrap test at each crossover, or soften the 'more than halves'/'not matched' wording. This is load-bearing for RQ2 and for the 'richness beats quantity' conclusion.

minor comments (4)

[Appendix A; §3.1] The text states a 32-tag POS inventory with 31 tags in training, but Table 6 appears to list only 30 rows. Please reconcile the count.
[§3.4, Figure 4] The text says the clitic ga occurs with six distinct grammatical glosses, but Figure 4 enumerates only five (gen, nom, foc, q, pur) from the four POS contexts shown. The remaining contexts are described in prose; please clarify the count and the sixth reading.
[§3.2] The term 'position span-F1' is used without definition. Please define it and distinguish it from boundary-level F1 and token-level span F1.
[§3.7] The log-odds analysis that supports the 'near-constant multiplier ≈1.8' claim is only described verbally. Since it is used to reconcile the two readings of Figure 7, consider including the regression details or per-budget log-odds values in the appendix.

Circularity Check

0 steps flagged

No load-bearing circularity; the POS-tier claim is an independent controlled ablation, with the gold-POS caveat explicitly stated.

full rationale

The claimed derivation—gold POS improves grammatical glossing and the gain grows as data shrink—is not equivalent to its inputs. The POS tier is a separate annotation layer from the gloss tier; the paper's own ga analysis (§3.4) shows that a single POS tag (cp) is compatible with multiple glosses (gen, nom), so gloss labels are not read off POS tags by construction. The Gold-POS improvement is a measured outcome of a controlled feature ablation (Tables 2–3, per-seed McNemar tests), and the pipeline condition (predicted POS) actually fails to realize the gain (+0.40±0.77, non-significant), which is the opposite of what a circular construction would produce. The 'data multiplier' learning curves use gold POS at test time, but the paper flags this limitation explicitly in §3.7 ('These curves use gold POS and thus speak to the assisted-annotation scenario; a pipeline analogue would retrain the tagger at each size — future work.') and in §5, presenting the result as relevant to the assisted scenario; the fully automatic pipeline analogue is left as future work. That is a scope/validity caveat, not a circular step. Self-citations (Shimoji 2011, 2021, 2025) supply corpus provenance, workflow framing, and a prior proof-of-concept, but the paper's quantitative conclusions come from the experiments reported here, not from those citations; no uniqueness theorem or ansatz is imported to forbid alternatives. Overall, no circular step is identifiable; at most there is minor self-citation in the framing, hence score 1.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The claims rest on an author-built corpus treated as gold-standard ground truth, model hyperparameters chosen by hand (though held constant across conditions), and a single-language case-study design. No new theoretical entities are introduced.

free parameters (4)

BiLSTM hidden size = 128
Chosen by hand; not swept. Not load-bearing for the main claim since it is identical across all compared conditions.
Morpheme embedding dim = 64
Chosen by hand; not swept. Identical across conditions.
POS embedding dim = 16
Chosen by hand; not swept. Could theoretically affect the POS gain size, but not the qualitative direction.
Model checkpoints = best-dev selection
Best-dev selection is a standard but nonetheless fitted choice; the paper does not report robustness to this criterion.

axioms (3)

domain assumption The annotation produced by the single author (Michinori Shimoji) is internally consistent and reflects the 'gold' description of Irabu.
The corpus and all tags were produced by the author. No inter-annotator agreement is reported. The experiments treat this annotation as ground truth, so divergent analyses by other linguists would change the numbers.
domain assumption The POS tier carries information that is not already implicit in the learned morpheme representations.
The measured gain under Gold POS supports this empirically, but the possibility that a larger or different model would not need an explicit POS tier is acknowledged as future work.
domain assumption Training/validation/test splits at the utterance level prevent data leakage between the tagger and glossing model.
The paper states splits are at utterance level; the Stage-2 tool has a 'leak guard' for fine-tuning. Leakage between the tagger and glossing training is not explicitly analyzed.

pith-pipeline@v1.3.0-alltime-deepseek · 18919 in / 6166 out tokens · 45378 ms · 2026-08-02T05:19:12.495679+00:00 · methodology

0 comments

read the original abstract

Discourse data are the primary empirical basis of grammar writing in field linguistics, but producing interlinearized text is notoriously expensive - on the order of one hour of work per minute of recording. For endangered languages, where the time remaining to verify analyses with native speakers is itself limited, automating parts of the interlinearization workflow has direct documentary value. We implement a full neural annotation pipeline (morpheme segmentation, POS tagging, glossing) for Irabu Ryukyuan using deliberately small, transparent BiLSTM-CRF models, and evaluate it under a realistic hard constraint: approximately one hour of fully annotated discourse as the entire supervised resource. Two factors of the annotation itself are manipulated: its richness (with or without a POS tier) and its quantity (training budgets from 6 to 47 minutes). Gold POS improves grammatical glossing by +4.4 (SD 0.7) points (significant in all 5 seeds), and the gain grows as data shrink (+11.6 points at a quarter of the data); a POS tier more than halves the amount of glossed data needed to reach a given accuracy. In a fully automatic pipeline this gain is not yet realized: the tagger still errs on 12% of morphemes, and an incorrect POS misleads the glossing model more than no POS at all. The value is latent rather than lost: degrading gold POS with controlled noise shows the gain returning as tagger accuracy rises, with break-even near our tagger's current 88% and +1.6 to +3.2 points recovered at 92-96%. We conclude with a concrete recommendation for documentation practice: annotate quadrilinearly - text, POS, gloss, translation.

discussion (0)

Reference graph

Works this paper leans on

31 extracted references · 6 linked inside Pith

[1]

Aycock, S., Stap, D., Wu, D., Monz, C., and Sima’an, K. (2025). Can LLMs really learn to translate a low-resource language from one grammar book? arXiv:2409.19151. Barriga Martínez, D., Mijangos, V., and Gutierrez-Vasques, X. (2021). Automatic interlinear glossing for Otomi language. In Proceedings of the First Workshop on Natural Language Pro- cessing fo...

Pith/arXiv arXiv 2025
[2]

and Liu, D

Elsner, M. and Liu, D. (2025). Prompt and circumstance: A word-by-word LLM prompting approach to interlinear glossing for low-resource languages. arXiv:2502.09778

Pith/arXiv arXiv 2025
[3]

Ginn, M., Moeller, S., Palmer, A., Stacey, A., Nicolai, G., Hulden, M., and Silfverberg, M. (2023). Findings of the SIGMOR- PHON 2023 shared task on interlinear glossing. InProceedings of the 20th SIGMORPHON Workshop on Computational Re- search in Phonetics, Phonology, and Morphology, pages 186– 201, Toronto, Canada. Association for Computational Linguis- tics

2023
[4]

Ginn, M., Palmer, A., and Hulden, M. (2024a). Can we teach language models to gloss endangered languages? arXiv:2406.18895

Pith/arXiv arXiv
[5]

Ginn, M., Tjuatja, L., He, T., Rice, E., Neubig, G., Palmer, A., and Levin, L. (2024b). GlossLM: A massively multilin- gual corpus and pretrained model for interlinear glossed text. arXiv:2403.06399

Pith/arXiv arXiv
[6]

Girrbach, L. (2023). Tü-CL at SIGMORPHON 2023: Straight- through gradient estimation for hard attention. InProceedings of the 20th SIGMORPHON Workshop on Computational Re- search in Phonetics, Phonology, and Morphology, pages 151– 165, Toronto, Canada. Association for Computational Linguis- tics

2023
[7]

S., and Fraser, A

Hangya, V., Saadi, H. S., and Fraser, A. (2022). Improving low- resourcelanguagesinpre-trainedmultilinguallanguagemodels. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11993–12006, Abu Dhabi, United Arab Emirates. Association for Computational Linguis- tics

2022
[8]

Databases for endangered languages: Japonic

Hidaka, M. (2026). Zenkoku h¯ ogen bunp¯ o jiten purojekuto no torikumi to tenkai [The Dictionary of Japanese Dialect Gram- mar project: Activities and developments]. InNihongo Gakkai 2026-nendo shunki taikai yok¯ osh¯ u [Proceedings of the Spring 2026 Meeting of the Society for Japanese Linguistics], pages 221–226. (In Japanese) Symposium “Databases for ...

2026
[9]

Himmelmann, N. P. (1998). Documentary and descriptive linguis- tics.Linguistics, 36(1):161–195

1998
[10]

and Schmidhuber, J

Hochreiter, S. and Schmidhuber, J. (1997). Long short-term mem- ory.Neural Computation, 9(8):1735–1780

1997
[11]

Lafferty, J., McCallum, A., and Pereira, F. C. N. (2001). Condi- tional random fields: Probabilistic models for segmenting and labelingsequencedata. InProceedings of the 18th International Conference on Machine Learning (ICML), pages 282–289

2001
[12]

Dyer, C. (2016). Neural architectures for named entity recog- nition. InProceedings of the 2016 Conference of the North American Chapter of the Association for Computational Lin- guistics: Human Language Technologies, pages 260–270, San

2016
[13]

Lehmann, C. (2004). Interlinear morphemic glossing. In Booij, G., Lehmann, C., Mugdan, J., and Skopeteas, S., editors, Morphologie / Morphology: Ein internationales Handbuch zur Flexion und Wortbildung, volume 2, pages 1834–1857. De Gruyter Mouton, Berlin

2004
[14]

McMillan-Major, A. (2020). Automating gloss generation in in- terlinear glossed text. In Ettinger, A., Jarosz, G., and Pater, J., editors,Proceedings of the Society for Computation in Linguis- tics 2020, pages 355–366, New York, New York. Association for Computational Linguistics

2020
[15]

Integratingautomatictranscriptionintothelanguage documentation workflow: Experiments with Na data and the Persephone toolkit.Language Documentation & Conservation, 12:393–429

Michaud, A., Adams, O., Cohn, T.A., Neubig, G., andGuillaume, S.(2018). Integratingautomatictranscriptionintothelanguage documentation workflow: Experiments with Na data and the Persephone toolkit.Language Documentation & Conservation, 12:393–429

2018
[16]

Miyagawa, S. (2026). Amami-gunt¯ o Yoron-t¯ o h¯ ogen no kikai hon’yaku e no LLM-RAG no ¯ oy¯ o [Applying LLM-based retrieval-augmented generation to machine translation of the Yoron dialect of the Amami Islands].IEICE Technical Report, 125(366, NLC2025-15):19–24. (In Japanese)

2026
[17]

and Hulden, M

Moeller, S. and Hulden, M. (2018). Automatic glossing in a low- resource setting for language documentation. In Klavans, J. L., editor,Proceedings of the Workshop on Computational Mod- eling of Polysynthetic Languages, pages 84–93, Santa Fe, New

2018
[18]

Association for Computational Linguistics

Mexico, USA. Association for Computational Linguistics. National Institute for Japanese Language and Linguistics (2016– 2025). Corpus of Japanese dialects (COJADS).https:// www2.ninjal.ac.jp/cojads/. Transcriptions with standard- Japanese translations; no morpheme-level annotation

2016
[19]

Ogueji, K., Zhu, Y., and Lin, J. (2021). Small data? No prob- lem! Exploringtheviabilityofpretrainedmultilinguallanguage models for low-resourced languages. InProceedings of the 1st Workshop on Multilingual Representation Learning, pages116– 126, Punta Cana, Dominican Republic. Association for Com- putational Linguistics. 13

2021
[20]

Palmer, A., Moon, T., and Baldridge, J. (2009). Evaluating au- tomation strategies in language documentation. In Ringger, E., Haertel, R., and Tomanek, K., editors,Proceedings of the NAACL HLT 2009 Workshop on Active Learning for Natural Language Processing, pages 36–44, Boulder, Colorado. Associ- ation for Computational Linguistics

2009
[21]

Ruokolainen, T., Kohonen, O., Virpioja, S., and Kurimo, M. (2013). Supervised morphological segmentation in a low- resource learning setting using conditional random fields. In Proceedings of the Seventeenth Conference on Computational Natural Language Learning, pages 29–37, Sofia, Bulgaria. As- sociation for Computational Linguistics

2013
[22]

Seifart, F., Evans, N., Hammarström, H., and Levinson, S. C. (2018). Language documentation twenty-five years on.Lan- guage, 94(4):e324–e345

2018
[23]

(2008).A Grammar of Irabu, a Southern Ryukyuan Language

Shimoji, M. (2008).A Grammar of Irabu, a Southern Ryukyuan Language. PhD thesis, Australian National University

2008
[24]

Shimoji, M. (2011). Bunp¯ o kijutsu ni okeru tekisuto no j¯ uy¯ osei [The importance of texts in grammar writing].Nihongogaku, 30(6):46–59. (In Japanese)

2011
[25]

(2017).A Grammar of Irabu: A Southern Ryukyuan Language

Shimoji, M. (2017).A Grammar of Irabu: A Southern Ryukyuan Language. Kyushu University Press, Fukuoka

2017
[26]

Shimoji, M. (2021). Guram¯ a raitingu: H¯ ogen no kijutsu bunp¯ o o kaku gaido [Grammar writing: A guide to writing a descriptive grammar of a dialect].Nihongo Bunp¯ o [Japanese Grammar], 21(2):136–151. (In Japanese)

2021
[27]

Shimoji, M. (2025). Teishigen gengo ni okeru tekisuto jid¯ o kaiseki to gurosu-zuke no jid¯ oka: Irabujima h¯ ogen no k¯ esu sutadi [Au- tomating text analysis and glossing for a low-resource language: A case study of Irabu Ryukyuan].Ry¯ uky¯ u no h¯ ogen [Ryukyuan dialects], 49:149–168. (In Japanese)

2025
[28]

(2013).Miyako Irabu h¯ ogen jiten [A dictionary of the Irabu dialect of Miyako]

Tomihama, S. (2013).Miyako Irabu h¯ ogen jiten [A dictionary of the Irabu dialect of Miyako]. Okinawa Times, Naha. (In Japanese)

2013
[29]

N., Kaiser, L., and Polosukhin, I

Gomez, A. N., Kaiser, L., and Polosukhin, I. (2017). Attention is all you need. arXiv:1706.03762

Pith/arXiv arXiv 2017
[30]

Yang, C., Nicolai, G., and Silfverberg, M. (2024). Em- bedded translations for low-resource automated glossing. arXiv:2403.08189

Pith/arXiv arXiv 2024
[31]

Zhao, X., Ozaki, S., Anastasopoulos, A., Neubig, G., and Levin, L. (2020). Automatic interlinear glossing for under-resourced languages leveraging translations. InProceedings of the 28th International Conference on Computational Linguistics, pages 5397–5408, Barcelona, Spain (Online). International Commit- tee on Computational Linguistics. 14 A POS Tag In...

2020

[1] [1]

Aycock, S., Stap, D., Wu, D., Monz, C., and Sima’an, K. (2025). Can LLMs really learn to translate a low-resource language from one grammar book? arXiv:2409.19151. Barriga Martínez, D., Mijangos, V., and Gutierrez-Vasques, X. (2021). Automatic interlinear glossing for Otomi language. In Proceedings of the First Workshop on Natural Language Pro- cessing fo...

Pith/arXiv arXiv 2025

[2] [2]

and Liu, D

Elsner, M. and Liu, D. (2025). Prompt and circumstance: A word-by-word LLM prompting approach to interlinear glossing for low-resource languages. arXiv:2502.09778

Pith/arXiv arXiv 2025

[3] [3]

Ginn, M., Moeller, S., Palmer, A., Stacey, A., Nicolai, G., Hulden, M., and Silfverberg, M. (2023). Findings of the SIGMOR- PHON 2023 shared task on interlinear glossing. InProceedings of the 20th SIGMORPHON Workshop on Computational Re- search in Phonetics, Phonology, and Morphology, pages 186– 201, Toronto, Canada. Association for Computational Linguis- tics

2023

[4] [4]

Ginn, M., Palmer, A., and Hulden, M. (2024a). Can we teach language models to gloss endangered languages? arXiv:2406.18895

Pith/arXiv arXiv

[5] [5]

Ginn, M., Tjuatja, L., He, T., Rice, E., Neubig, G., Palmer, A., and Levin, L. (2024b). GlossLM: A massively multilin- gual corpus and pretrained model for interlinear glossed text. arXiv:2403.06399

Pith/arXiv arXiv

[6] [6]

Girrbach, L. (2023). Tü-CL at SIGMORPHON 2023: Straight- through gradient estimation for hard attention. InProceedings of the 20th SIGMORPHON Workshop on Computational Re- search in Phonetics, Phonology, and Morphology, pages 151– 165, Toronto, Canada. Association for Computational Linguis- tics

2023

[7] [7]

S., and Fraser, A

Hangya, V., Saadi, H. S., and Fraser, A. (2022). Improving low- resourcelanguagesinpre-trainedmultilinguallanguagemodels. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11993–12006, Abu Dhabi, United Arab Emirates. Association for Computational Linguis- tics

2022

[8] [8]

Databases for endangered languages: Japonic

Hidaka, M. (2026). Zenkoku h¯ ogen bunp¯ o jiten purojekuto no torikumi to tenkai [The Dictionary of Japanese Dialect Gram- mar project: Activities and developments]. InNihongo Gakkai 2026-nendo shunki taikai yok¯ osh¯ u [Proceedings of the Spring 2026 Meeting of the Society for Japanese Linguistics], pages 221–226. (In Japanese) Symposium “Databases for ...

2026

[9] [9]

Himmelmann, N. P. (1998). Documentary and descriptive linguis- tics.Linguistics, 36(1):161–195

1998

[10] [10]

and Schmidhuber, J

Hochreiter, S. and Schmidhuber, J. (1997). Long short-term mem- ory.Neural Computation, 9(8):1735–1780

1997

[11] [11]

Lafferty, J., McCallum, A., and Pereira, F. C. N. (2001). Condi- tional random fields: Probabilistic models for segmenting and labelingsequencedata. InProceedings of the 18th International Conference on Machine Learning (ICML), pages 282–289

2001

[12] [12]

Dyer, C. (2016). Neural architectures for named entity recog- nition. InProceedings of the 2016 Conference of the North American Chapter of the Association for Computational Lin- guistics: Human Language Technologies, pages 260–270, San

2016

[13] [13]

Lehmann, C. (2004). Interlinear morphemic glossing. In Booij, G., Lehmann, C., Mugdan, J., and Skopeteas, S., editors, Morphologie / Morphology: Ein internationales Handbuch zur Flexion und Wortbildung, volume 2, pages 1834–1857. De Gruyter Mouton, Berlin

2004

[14] [14]

McMillan-Major, A. (2020). Automating gloss generation in in- terlinear glossed text. In Ettinger, A., Jarosz, G., and Pater, J., editors,Proceedings of the Society for Computation in Linguis- tics 2020, pages 355–366, New York, New York. Association for Computational Linguistics

2020

[15] [15]

Integratingautomatictranscriptionintothelanguage documentation workflow: Experiments with Na data and the Persephone toolkit.Language Documentation & Conservation, 12:393–429

Michaud, A., Adams, O., Cohn, T.A., Neubig, G., andGuillaume, S.(2018). Integratingautomatictranscriptionintothelanguage documentation workflow: Experiments with Na data and the Persephone toolkit.Language Documentation & Conservation, 12:393–429

2018

[16] [16]

Miyagawa, S. (2026). Amami-gunt¯ o Yoron-t¯ o h¯ ogen no kikai hon’yaku e no LLM-RAG no ¯ oy¯ o [Applying LLM-based retrieval-augmented generation to machine translation of the Yoron dialect of the Amami Islands].IEICE Technical Report, 125(366, NLC2025-15):19–24. (In Japanese)

2026

[17] [17]

and Hulden, M

Moeller, S. and Hulden, M. (2018). Automatic glossing in a low- resource setting for language documentation. In Klavans, J. L., editor,Proceedings of the Workshop on Computational Mod- eling of Polysynthetic Languages, pages 84–93, Santa Fe, New

2018

[18] [18]

Association for Computational Linguistics

Mexico, USA. Association for Computational Linguistics. National Institute for Japanese Language and Linguistics (2016– 2025). Corpus of Japanese dialects (COJADS).https:// www2.ninjal.ac.jp/cojads/. Transcriptions with standard- Japanese translations; no morpheme-level annotation

2016

[19] [19]

Ogueji, K., Zhu, Y., and Lin, J. (2021). Small data? No prob- lem! Exploringtheviabilityofpretrainedmultilinguallanguage models for low-resourced languages. InProceedings of the 1st Workshop on Multilingual Representation Learning, pages116– 126, Punta Cana, Dominican Republic. Association for Com- putational Linguistics. 13

2021

[20] [20]

Palmer, A., Moon, T., and Baldridge, J. (2009). Evaluating au- tomation strategies in language documentation. In Ringger, E., Haertel, R., and Tomanek, K., editors,Proceedings of the NAACL HLT 2009 Workshop on Active Learning for Natural Language Processing, pages 36–44, Boulder, Colorado. Associ- ation for Computational Linguistics

2009

[21] [21]

Ruokolainen, T., Kohonen, O., Virpioja, S., and Kurimo, M. (2013). Supervised morphological segmentation in a low- resource learning setting using conditional random fields. In Proceedings of the Seventeenth Conference on Computational Natural Language Learning, pages 29–37, Sofia, Bulgaria. As- sociation for Computational Linguistics

2013

[22] [22]

Seifart, F., Evans, N., Hammarström, H., and Levinson, S. C. (2018). Language documentation twenty-five years on.Lan- guage, 94(4):e324–e345

2018

[23] [23]

(2008).A Grammar of Irabu, a Southern Ryukyuan Language

Shimoji, M. (2008).A Grammar of Irabu, a Southern Ryukyuan Language. PhD thesis, Australian National University

2008

[24] [24]

Shimoji, M. (2011). Bunp¯ o kijutsu ni okeru tekisuto no j¯ uy¯ osei [The importance of texts in grammar writing].Nihongogaku, 30(6):46–59. (In Japanese)

2011

[25] [25]

(2017).A Grammar of Irabu: A Southern Ryukyuan Language

Shimoji, M. (2017).A Grammar of Irabu: A Southern Ryukyuan Language. Kyushu University Press, Fukuoka

2017

[26] [26]

Shimoji, M. (2021). Guram¯ a raitingu: H¯ ogen no kijutsu bunp¯ o o kaku gaido [Grammar writing: A guide to writing a descriptive grammar of a dialect].Nihongo Bunp¯ o [Japanese Grammar], 21(2):136–151. (In Japanese)

2021

[27] [27]

Shimoji, M. (2025). Teishigen gengo ni okeru tekisuto jid¯ o kaiseki to gurosu-zuke no jid¯ oka: Irabujima h¯ ogen no k¯ esu sutadi [Au- tomating text analysis and glossing for a low-resource language: A case study of Irabu Ryukyuan].Ry¯ uky¯ u no h¯ ogen [Ryukyuan dialects], 49:149–168. (In Japanese)

2025

[28] [28]

(2013).Miyako Irabu h¯ ogen jiten [A dictionary of the Irabu dialect of Miyako]

Tomihama, S. (2013).Miyako Irabu h¯ ogen jiten [A dictionary of the Irabu dialect of Miyako]. Okinawa Times, Naha. (In Japanese)

2013

[29] [29]

N., Kaiser, L., and Polosukhin, I

Gomez, A. N., Kaiser, L., and Polosukhin, I. (2017). Attention is all you need. arXiv:1706.03762

Pith/arXiv arXiv 2017

[30] [30]

Yang, C., Nicolai, G., and Silfverberg, M. (2024). Em- bedded translations for low-resource automated glossing. arXiv:2403.08189

Pith/arXiv arXiv 2024

[31] [31]

Zhao, X., Ozaki, S., Anastasopoulos, A., Neubig, G., and Levin, L. (2020). Automatic interlinear glossing for under-resourced languages leveraging translations. InProceedings of the 28th International Conference on Computational Linguistics, pages 5397–5408, Barcelona, Spain (Online). International Commit- tee on Computational Linguistics. 14 A POS Tag In...

2020