Pith. sign in

REVIEW 6 major objections 6 minor 40 references

Building a Functional Machine Translation Corpus for Kpelle

T0 review · 6 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read First English-Kpelle corpus reaches BLEU 30

desk verdict A useful new Kpelle-English corpus, but the paper's own numbers are inconsistent enough that the headline BLEU scores should be treated as provisional until the dataset is cleaned and re-described. read the letter →

arxiv 2505.18905 v1 pith:T25WVEFI submitted 2025-05-24 cs.CL

Kweku Andoh Yamoah , Jackson Weako , Emmanuel J. Dorley This is my paper · ORCID
classification cs.CL
keywords Kpellemachinetranslationlow-resourcelanguagesparallelcorpusNLLBMandeLiberiatone-markedorthography
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces the first publicly available English-Kpelle parallel corpus, built for machine translation and language learning. The dataset contains 3,234 translation pairs drawn from everyday conversation, religious texts, and educational materials, normalized into tone-marked Latin orthography. The authors claim that fine-tuning the NLLB multilingual translation model on this corpus yields a BLEU score of about 30 for Kpelle-to-English and about 24 for English-to-Kpelle, and that moving from the smaller Version 1 to the augmented Version 2 improves the Kpelle-to-English direction. A sympathetic reading: if these results hold, Kpelle gains a reusable resource for translation and other NLP tasks, and low-resource Mande languages enter the same performance range as other African languages in NLLB-200.

What carries the argument

The load-bearing object is the dataset itself: 3,234 English-Kpelle translation pairs, 30,021 words total, 4,702 unique Kpelle words, and two corpus versions (V1 with 1,518 pairs and V2 with 2,005 pairs) that isolate the effect of adding data. The same NLLB model is fine-tuned on each version at 10k, 30k, and 60k steps, with evaluation by BLEU and chrF2++ (a character-based metric); the V1-to-V2 comparison in Kpelle-to-English translation is the experiment that carries the claim that the corpus is functional and that more data helps.

What would settle it

Randomly sample 200 held-out sentence pairs, have two additional native Kpelle speakers independently produce reference translations, and recompute BLEU for the released fine-tuned model against the new references; if the scores fall well below the reported range of 24–30, the performance is an artifact of the original reference set. A second check would be to renormalize the entire corpus under a documented orthographic standard and see whether fine-tuning results change materially.

Watch

Extended reading notes

Core claim

The central claim is that a carefully assembled bilingual dataset can make Kpelle machine translation work despite the language's low-resource status. Fine-tuning NLLB on Version 2 reaches a best BLEU of 30.28 for Kpelle-to-English at 60,000 training steps and 24.46 for English-to-Kpelle at 30,000 steps on Version 1; these scores are presented as consistent with NLLB-200's reported range for other African languages. The paper also claims that expanding the corpus from 1,518 to 2,005 translation pairs produces measurable gains in the Kpelle-to-English direction, demonstrating the value of data augmentation in a low-resource setting.

Load-bearing premise

The corpus's usefulness rests on the assumption that expert sentence alignments and the tone-marked Latin orthography are consistently accurate across all pairs; if translations are misaligned or tone marks are inconsistent, the reported BLEU scores are not a reliable signal of translation quality.

Editorial extensions

If this is right

  • Kpelle becomes a reproducible testbed for low-resource translation: released sentence pairs allow other teams to fine-tune and evaluate models without recollecting data.
  • Expanding a low-resource corpus by a few hundred in-domain pairs can move BLEU measurably in one translation direction, suggesting data curation is a high-leverage investment.
  • The same corpus can seed speech recognition, language modeling, and sentiment analysis because it provides normalized orthography with tone marks.
  • The reported Kpelle scores give future systems a concrete baseline to beat and a point of comparison against NLLB-200 languages such as Wolof, Luo, Yoruba, and Luganda.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the corpus added Guinean Kpelle variants, translation quality would probably shift; an experiment separating Liberian and Guinean sources could quantify how much the macro-language split costs current models.
  • The high number of Kpelle hapax legomena (2,714 words appearing once) suggests the reported BLEU scores may not reflect fluency on rare words; a human evaluation or a rare-word breakdown would test this.
  • Because the domain table shows Religion, Health, and Education are thinly covered, targeted expansion of those domains could improve scores more than a generic increase in pair count.
  • A useful extension would be to benchmark the same corpus on a second model architecture; divergence between models would reveal which errors come from the data and which from NLLB's inductive biases.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper introduces a new English–Kpelle parallel corpus, assembled from travel phrasebooks, religious texts, and educational materials, with expert translations and verification. The authors fine-tune Meta's NLLB model on two versions of the corpus and report BLEU scores of up to 30.28 for Kpelle-to-English and 24.46 for English-to-Kpelle, interpreting these as evidence that Kpelle can reach performance levels comparable to other low-resource African languages in NLLB-200. The paper also describes domain statistics, vocabulary analyses, and a roadmap for future expansion.

Significance. If the corpus and evaluation are validated, this is a meaningful contribution: Kpelle is a major Liberian language with very few digital resources, and a publicly released parallel corpus with expert involvement would support machine translation and downstream NLP work. The authors are appropriately explicit about some limitations, and the dataset is placed on Hugging Face. However, the significance is conditional on resolving internal count inconsistencies, verifying orthographic and tonal consistency, and strengthening the evaluation design; the current experiments do not yet establish the headline BLEU claims.

major comments (6)
  1. [Section 5.1, Section 6.1, Table 2] The dataset size is reported inconsistently: Section 5.1 says 3,234 translation pairs, footnote 2 says Version 2 extends to 2,005 pairs, Section 6.1 gives 2,202 Kpelle and 2,167 English sentences for V2, and Table 2's domain counts sum to 2,005. These numbers cannot all be correct under a one-sentence-per-pair definition. Please define the exact unit (sentence pair versus multi-sentence entry) and reconcile all counts, since the abstract's 'over 2,000 pairs' and Contribution (a)'s '3,234 translation pairs' are in direct tension.
  2. [Section 4.2.2, Section 5.5, Section 3.1.4] The paper first states that diacritical marks were standardized to represent tonal variations accurately, but Section 5.5 admits 'spelling and tone-marking variations throughout the dataset.' Because Section 3.1.4 shows that tone is lexically contrastive (e.g., lá 'mouth' vs. là 'if'), any unresolved tone variation in the reference data can invalidate BLEU estimates and downstream use. Report a quantitative consistency check, such as inter-annotator agreement on tone marking or a normalized-token audit, and, if residual variation remains, assess its effect on the held-out test set.
  3. [Section 6.1, Table 3] No zero-shot NLLB scores are reported on the Kpelle held-out sets. Without these baselines, the reported BLEU gains cannot be attributed to fine-tuning or to the corpus; the model may already produce comparable scores before any training on Kpelle. Please add zero-shot NLLB (and ideally another untrained baseline) evaluated on the same test splits.
  4. [Section 6.1, Section 4] The comparison between V1 and V2 is presented as evidence for the benefit of 'data augmentation,' but the augmentation method is never described, and the two versions differ in size and likely in source composition. This confound makes the V1/V2 comparison uninterpretable as an augmentation study. Define the augmentation procedure explicitly and, if possible, ablate it while holding the underlying source data fixed.
  5. [Section 6.3, Table 4] The cross-lingual BLEU comparisons are not valid because the scores come from different test sets. The Kpelle numbers are computed on a small in-domain held-out split of this corpus, whereas the cited NLLB-200, M2M-100, and MMTAfrica scores are based on different benchmark test sets. Direct statements such as 'surpasses NLLB-200's lower-bound performances' should be removed or explicitly reframed as non-comparable contextual references; otherwise the claim that Kpelle reaches NLLB-200-level performance is unsupported.
  6. [Section 6.2, Table 3] The experiments report a single run with no confidence intervals or multiple seeds, and the headline BLEU of 30.28 is the maximum over three step counts. Please report variance across at least three random seeds and state the model-selection rule (e.g., choose the step count on a validation split) before selecting the best result.
minor comments (6)
  1. [References] The TangaleNLP reference contains what appears to be a typo ('po tangle'); please correct the title and provide a persistent DOI or URL for the dataset repository.
  2. [Section 5.3] The sentence 'Even though we remove common stop words' is confusing because the listed frequent English words are content words, not stop words; please describe the stop-word list used and align the Kpelle function-word examples with the actual top-frequency list.
  3. [Table 3] The row label 'NLLB' is ambiguous; use 'NLLB V1' and 'NLLB V2' to distinguish the fine-tuned versions from the un-fine-tuned baseline model.
  4. [Figure 3] Figure 3(b) shows a single translation example; please state whether it was selected as representative or as a best-case output, and consider adding a few examples from different domains.
  5. [Section 5.5] The paragraph repeats the same point about underrepresented domains twice; please condense to one statement.
  6. [Section 3.1.3] The emphatic particle appears as 'b´ e' with a stray accent; please use a consistent diacritic encoding throughout the paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the BLEU results come from held-out sacreBLEU evaluation after fine-tuning NLLB, and the only self-citation is a non-load-bearing tone-marking reference.

full rationale

This paper does not claim a first-principles derivation; it constructs a dataset and benchmarks it. The central quantitative claims (BLEU ~30 kpe_Latn→eng_Latn and ~24 eng_Latn→kpe_Latn) are produced by fine-tuning NLLB on training splits and scoring with sacreBLEU on held-out test splits (Section 6.1, Table 3). The evaluation is therefore not circular with respect to the test set by construction. The V1 versus V2 comparison is a simple more-data experiment; although it is confounded by differing corpus sizes and sources, it does not reduce to an identity between input and output. The only citation to the authors' own prior work is [Weako 2024] in Table 1, used to illustrate tone-marking conventions; this citation is not load-bearing for the dataset creation, alignment, normalization, or the reported BLEU scores. Admitted internal inconsistencies, such as the discrepancy between '3,234 translation pairs' in Section 5.1 and the V2 counts of 2,005 pairs or 2,202/2,167 sentences, and the acknowledged 'spelling and tone-marking variations throughout the dataset' in Section 5.5, are data-quality and reproducibility concerns rather than circularity: they do not make the reported metrics equivalent to the input by definition. The claim of being 'the first publicly available English-Kpelle parallel dataset' is an existence claim, not a self-referential derivation. No load-bearing step in the paper's chain reduces a prediction to a fitted parameter, a self-citation, or a redefined known result. The appropriate finding is no significant circularity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical entities. Its claims rest on data quality and evaluation assumptions that are not externally verified.

free parameters (1)
  • Fine-tuning step count (best-of-three selection) = 60k steps (BLEU 30.28 for kpe->eng)
    The headline BLEU of about 30 is the maximum over 10k, 30k, and 60k steps, so the reported result selects the best of three training lengths rather than a single pre-specified configuration.
assumptions (4)
  • domain assumption The expert translations and sentence alignments are accurate and idiomatic.
    The dataset's quality rests on two named experts (Section 4.1.2) without inter-annotator agreement or external validation.
  • domain assumption The Latin-based orthography was normalized consistently, preserving tonal distinctions.
    Kpelle has no standardized orthography (Section 3.1.5); the paper asserts standardization in Section 4.2.2 but provides no consistency checks.
  • domain assumption The NLLB-200 model's kpe_Latn language code corresponds to the same Liberian Kpelle variant used in the corpus.
    The paper focuses on Liberian Kpelle (Section 3.1) but uses NLLB's kpe_Latn code, which may cover both Liberian and Guinean variants; the alignment is not discussed.
  • standard math SacreBLEU evaluated on the held-out 10% test set is a reliable measure of translation quality for Kpelle.
    The BLEU metric is standard, but with only about 200 test sentences the score has high variance; the paper reports no confidence intervals.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Building a Functional Machine Translation Corpus for Kpelle." pith.science (2026). https://pith.science/paper/T25WVEFI

@misc{pith2026250518905,
  author       = {Pith},
  title        = {Pith review of: Building a Functional Machine Translation Corpus for Kpelle},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T25WVEFI}},
  note         = {Machine review of arXiv:2505.18905}
}
read the original abstract

In this paper, we introduce the first publicly available English-Kpelle dataset for machine translation, comprising over 2000 sentence pairs drawn from everyday communication, religious texts, and educational materials. By fine-tuning Meta's No Language Left Behind(NLLB) model on two versions of the dataset, we achieved BLEU scores of up to 30 in the Kpelle-to-English direction, demonstrating the benefits of data augmentation. Our findings align with NLLB-200 benchmarks on other African languages, underscoring Kpelle's potential for competitive performance despite its low-resource status. Beyond machine translation, this dataset enables broader NLP tasks, including speech recognition and language modelling. We conclude with a roadmap for future dataset expansion, emphasizing orthographic consistency, community-driven validation, and interdisciplinary collaboration to advance inclusive language technology development for Kpelle and other low-resourced Mande languages.

Figures

Figures reproduced from arXiv: 2505.18905 by the authors.

Figure 1
Figure 1. Overview of Liberian language family under the Niger-Congo Branch. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Sentence length distributions for English (top) [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. NLLB-200 fine-tuning with Kpelle: (a) Model adaptation for bidirectional translation, and (b) a sample [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 27 canonical work pages

  1. [1]

    Mark Abadi. 2018. https://www.businessinsider.com/travel-language-phrases-to-learn-2018-6 I've been to 25 countries and i can tell you there are only 11 phrases you need to get by anywhere

    work page 2018

  2. [2]

    David Ifeoluwa Adelani, Jesujoba Oluwadara Alabi, Angela Fan, Julia Kreutzer, Xiaoyu Shen, Machel Reid, Dana Ruiter, Dietrich Klakow, Peter Nabende, Ernie Chang, Tajuddeen Gwadabe, Freshia Sackey, Bonaventure F. P. Dossou, Chris Chinenye Emezue, Colin Leong, Michael Beukman, Shamsuddeen Hassan Muhammad, Guyo Dub Jarso, Oreen Yousuf, Andre Niyongabo Rubung...

    work page Pith review arXiv doi:10.48550/arxiv.2205.02022 2022

  3. [3]

    University of Wisconsin-Madison Students in African 671. 2019. https://wisc.pb.unizin.org/lctlresources/chapter/kpelle-history-and-brief-intro-lesson-on-syllabary-and-alphabet/ Kpelle- history and brief intro. Lesson on syllabary and alphabet . Publisher: Pressbooks

    work page 2019

  4. [4]

    Emmanuel Agyei, Xiaoling Zhang, Stephen Bannerman, Ama Bonuah Quaye, Sophyani Banaamwini Yussi, and Victor Kwaku Agbesi. 2024. https://doi.org/10.1007/s10791-024-09451-8 Low resource twi-english parallel corpus for machine translation in multiple domains (twi-2-eng) . Deleted Journal, 27

    doi:10.1007/s10791-024-09451-8 2024

  5. [5]

    Adewale Akinfaderin. 2020. https://doi.org/10.18653/v1/2020.winlp-1.38 H ausa MT v1.0: Towards E nglish -- H ausa neural machine translation . In Proceedings of the Fourth Widening Natural Language Processing Workshop, pages 144--147, Seattle, USA. Association for Computational Linguistics

    work page doi:10.18653/v1/2020.winlp-1.38 2020

  6. [6]

    Asamoah Owusu, A

    D. Asamoah Owusu, A. Korsah, B. Quartey, S. Nwolley Jnr., D. Sampah, D. Adjepon-Yamoah, and L. Omane Boateng. 2022. Github - ashesi-org/financial-inclusion-speech-dataset: A speech dataset to support financial inclusion created by ashesi university and nokwary technologies with funding from lacuna fund. https://github.com/Ashesi-Org/Financial-Inclusion-Sp...

    work page 2022

  7. [7]

    Gebremeskel , and Abel Aregawi

    Asmelash Teka Hadgu , Gebrekirstos G. Gebremeskel , and Abel Aregawi . 2022. https://github.com/asmelashteka/HornMT Machine Translation Benchmark Dataset for Languages in the Horn of Africa . Original-date: 2021-12-05T14:04:38Z

    work page 2022

  8. [8]

    Sara B. 2018. https://www.ef.edu/blog/language/13-important-phrases-know-second-language/ 13 important phrases to know in your second language

    work page 2018

Show all 40 references
  1. [9]

    Gideon George, Olubayo Adekanmbi, and Anthony Soronnadi. 2024. https://openreview.net/forum?id=xLNUQTWsS7 Tangale NLP : Building po tangle to english parallel corpora and machine translation of the tangle (tangale) language . In 5th Workshop on African Natural Language Processing

    2024

  2. [10]

    Naman Goyal, Cynthia Gao, Vishrav Chaudhary, Peng-Jen Chen, Guillaume Wenzek, Da Ju, Sanjana Krishnan, Marc'Aurelio Ranzato, Francisco Guzman, and Angela Fan. 2021. https://arxiv.org/abs/2106.03193 The flores-101 evaluation benchmark for low-resource and multilingual machine t...

    2021 arXiv

  3. [11]

    Heine and M

    B. Heine and M. Reh. 1984. https://books.google.com.gh/books?id=36YOAAAAYAAJ Grammaticalization and Reanalysis in African Languages . H. Buske

    1984

  4. [12]

    Maria Konoshenko. 2024. https://doi.org/10.30842/alp23065737193558583 Quotatives in guinean and liberian kpelle: A study of parallel bible corpora and non-biblical texts . Acta Linguistica Petropolitana, 19(3):558--583

    2024 doi

  5. [13]

    Maria Yu Konoshenko. 2008. https://api.semanticscholar.org/CorpusID:148565767 Tonal systems in three dialects of the kpelle language . Mandenkan

    2008

  6. [14]

    Taku Kudo and John Richardson. 2018. https://arxiv.org/abs/1808.06226 Sentencepiece: A simple and language independent subword tokenizer and detokenizer for neural text processing . CoRR, abs/1808.06226

    2018 arXiv

  7. [15]

    Siva Subrahamanyam Varma Kusampudi, Anudeep Chaluvadi, and Radhika Mamidi. 2021. https://aclanthology.org/2021.ranlp-1.85 Corpus creation and language identification in low-resource code-mixed T elugu- E nglish text . In Proceedings of the International Conference on Recent Ad...

    2021

  8. [16]

    Leidenfrost and John S

    Theodore E. Leidenfrost and John S. McKay. 2005. Kpelle-English Dictionary, with a Grammar Sketch and English-Kpelle Finder List. Language-Literacy-Literature and Bible Translation Center- Lutheran Church in Liberia, Totota

    2005

  9. [17]

    Lonely Planet Global Limited. 2018. https://premiki.si/wp-content/uploads/2018/05/accessible-travel-phrasebook-1.pdf 35 languages covered accessible travel phrasebook

    2018

  10. [18]

    Jean Maillard, Cynthia Gao, Elahe Kalbassi, Kaushik Ram Sadagopan, Vedanuj Goswami, Philipp Koehn, Angela Fan, and Francisco Guzman. 2023. https://doi.org/10.18653/v1/2023.acl-long.154 Small data, big impact: Leveraging minimal data for effective machine translation . In Proce...

    2023 doi

  11. [19]

    Joyce Nakatumba-Nabende, Claire Babirye, Peter Nabende, Jeremy Francis Tusubira, Jonathan Mukiibi, Eric Peter Wairagala, Chodrine Mutebi, Tobius Saul Bateesa, Alvin Nahabwe, Hewitt Tusiime, and Andrew Katumba. 2024. https://doi.org/10.1002/ail2.92 Building text and speech benc...

    2024 doi

  12. [20]

    Wilhelmina Nekoto, Vukosi Marivate, Tshinondiwa Matsila, Timi Fasubaa, Taiwo Fagbohungbe, Solomon Oluwole Akinola, Shamsuddeen Muhammad, Salomon Kabongo Kabenamualu, Salomey Osei, Freshia Sackey, Rubungo Andre Niyongabo, Ricky Macharm, Perez Ogayo, Orevaoghene Ahia, Musie Mere...

    2020

  13. [21]

    Vinh Van Nguyen, Ha Nguyen, Huong Thanh Le, Thai Phuong Nguyen, Tan Van Bui, Luan Nghia Pham, Anh Tuan Phan, Cong Hoang-Minh Nguyen, Viet Hong Tran, and Anh Huu Tran. 2022. https://aclanthology.org/2022.lrec-1.588 KC 4 MT : A high-quality corpus for multilingual machine transl...

    2022

  14. [22]

    Iroro Orife, Julia Kreutzer, Blessing Sibanda, Daniel Whitenack, Kathleen Siminyu, Laura Martinus, Jamiil Toure Ali, Jade Abbott, Vukosi Marivate, Salomon Kabongo, Musie Meressa, Espoir Murhabazi, Orevaoghene Ahia, Elan van Biljon, Arshath Ramkilowan, Adewale Akinfaderin, Alp ...

    2020 arXiv

  15. [23]

    Denis Paperno. 2014. https://doi.org/10.4000/mandenkan.563 Sample texts in beng . Mandenkan, pages 106--111

    2014 doi

  16. [24]

    Guilherme Penedo, Hynek Kydlíček, Vinko Sabolčec, Bettina Messmer, Negar Foroutan, Martin Jaggi, Leandro von Werra, and Thomas Wolf. 2024. https://doi.org/10.57967/hf/3744 Fineweb2: A sparkling update with 1000s of languages

    2024 doi

  17. [25]

    Olivia Christine Perez. 2022. https://www.gooverseas.com/blog/language-phrases-before-travel Helpful language phrases to learn before you travel | go overseas

    2022

  18. [26]

    Matt Post. 2018. https://www.aclweb.org/anthology/W18-6319 A call for clarity in reporting BLEU scores . In Proceedings of the Third Conference on Machine Translation: Research Papers, pages 186--191, Belgium, Brussels. Association for Computational Linguistics

    2018

  19. [27]

    Paul Kanmu Ricks. 2009. Kwaa Pa Kpelee-Woo Maa Kori(We Have Come to Learn Kpelle), first edition. Cuttington University

    2009

  20. [28]

    NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prang...

    2022 arXiv

  21. [29]

    Sharon V Thach. 1981. https://nla.gov.au/nla.cat-vn5443671 A Learner Directed Approach to Kpelle. A Handbook on Communication and Culture with Dialogs, Texts, Cultural Notes, Exercises, Drills and Instructions [microform] / Sharon V. Thach and Others . Distributed by ERIC Clea...

    1981

  22. [30]

    Thach, D.J

    S.V. Thach, D.J. Dwyer, and Michigan State University. African Studies Center. 1981. https://books.google.com.gh/books?id=B3oOAAAAYAAJ Kpelle, a Reference Handbook of Phonetics, Grammar, Lexicon and Learning Procedures . [Prepared] for the United States Peace Corps at the Afri...

    1981

  23. [31]

    Valentin Vydrin. 2018. https://doi.org/10.1093/acrefore/9780199384655.013.397 Mande languages

    2018

  24. [32]

    Valentin Vydrin, Jean-Jacques Meric, Kirill Maslinsky, Andrij Rovenchak, Allashera Auguste Tapo, Sebastien Diarra, Christopher Homan, Marco Zampieri, and Michael Leventhal. 2022. Machine learning dataset development for manding languages. url https://github.com/robotsmali-ai/datasets

    2022

  25. [33]

    Alexandra Vydrina. 2017. https://shs.hal.science/tel-03203594 A corpus-based description of Kakabe, a Western Mande language: prosody in grammar . Theses, Institut National des Langues et Civilisations Orientales

    2017

  26. [34]

    Barack Wanjawa, Lilian D. A. Wanzare, Florence Indede, Owen McOnyango, Edward Ombui, and Lawrence Muchemi. 2024. https://doi.org/10.7910/DVN/6N5V1K Kencorpus: Kenyan Languages Corpus

    2024 doi

  27. [35]

    Jackson Weako. 2024. https://doi.org/1005580.ingest.sentry.io/5983515 Libtralo kpelle keyboard help . Keyman.com

    2024

  28. [36]

    Contributors Wikivoyage. 2005. https://en.wikivoyage.org/wiki/Afrikaans_phrasebook West germanic language, spoken in south africa and namibia

    2005

  29. [37]

    Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. https://doi.org/10.18653/v1/N18-1101 A broad-coverage challenge corpus for sentence understanding through inference . In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computation...

    2018 doi

  30. [38]

    Liam G.-Staff Writer. 2017. https://onlineteachersuk.com/english-for-tourism-travel/ English for tourism: Essential uk travel phrases with examples

    2017

  31. [39]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  32. [40]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.