Pith. sign in

REVIEW 2 major objections 4 minor 63 references

Do We Really Need Fully Unsupervised Cross-Lingual Embeddings?

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Fully unsupervised cross-lingual word embeddings, even at their most robust, never outperform weakly supervised embeddings seeded with a few hundred translation pairs, and fail completely on 87 of 210 language pairs.

desk verdict A careful, large-scale negative result on unsupervised CLWEs whose class-level headline claim outruns its single-algorithm evidence. read the letter →

arxiv 1909.01638 v1 pith:GWQLFYYI submitted 2019-09-04 cs.CL

Ivan Vulić , Goran Glavaš , Roi Reichart , Anna Korhonen This is my paper · ORCID
classification cs.CL
keywords cross-lingualwordembeddingsunsupervisedalignmentbilinguallexiconinductionweaksupervisionself-learningpreprocessinglanguagediversityrobustness
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

The paper asks whether fully unsupervised cross-lingual word embeddings deliver on their main promise: usable alignment for distant and resource-poor language pairs. It dissects a projection-based CLWE system into three components—seed dictionary extraction, iterative self-learning, and pre/postprocessing—and compares configurations that differ only in where the seed comes from. Across 210 bilingual lexicon induction setups over 15 diverse languages, the fully unsupervised configuration never beats the same pipeline seeded with 500–1,000 translation pairs, and it collapses to near-zero performance on 87/210 pairs, concentrated on typologically distant languages. The conclusion is that 'no supervision at all' buys little in practice, and the field's real gains have come from the self-learning and preprocessing components, which work even better with a small seed dictionary.

What carries the argument

The load-bearing mechanism is a component-wise decomposition of projection-based CLWE into C1 (seed dictionary extraction), C2 (iterative self-learning with orthogonal SVD projection, mutual nearest neighbours, and dropout), and C3 (four preprocessing/postprocessing steps: unit-length normalization with mean centering, ZCA whitening, dewhitening, and symmetric re-weighting). VECMAP supplies C1 in fully unsupervised form by sorting monolingual similarity distributions and taking mutual nearest neighbours across languages. The paper's experimental design holds C2 and C3 fixed while varying only C1, replacing unsupervised seed extraction with a provided dictionary of 500, 1K, or 5K translation pairs. This isolation lets the authors attribute performance differences to the seed source rather than to the self-learning machinery.

What would settle it

Run any second unsupervised alignment method (for instance a GAN-based, Gromov-Wasserstein, or hyperalignment-style approach) through the same S1–S4 and self-learning pipeline on the released PanLex 210-pair dictionaries with the same fastText vectors, and check whether any language pair yields MRR above the 500-pair FULL+SL+SYM score or rescues one of the 87 near-zero pairs. A positive result on a single typologically distant pair would falsify the paper's universal claim; the same test on the English–Finnish and English–Hungarian domain-mismatch setups would test the domain-robustness claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that the only substantive difference between fully unsupervised and weakly supervised projection-based CLWE is the seed dictionary; once that difference is isolated, the unsupervised advantage disappears. Across 210 bilingual lexicon induction setups spanning 15 genealogically and typologically diverse languages, the VECMAP unsupervised model produces MRR at or below 0.01 in 87/210 setups, and in no setup does it outperform the weakly supervised configuration that uses the same self-learning procedure and the same S1–S4 preprocessing/postprocessing, seeded with 5K, 1K, or 500 translation pairs. The best weakly supervised configuration (FULL+SL+SYM) wins 177/210 setups at 1K seed pairs and never fails. Unsupervised alignment also fails under domain mismatch between the two monolingual corpora, succeeding only for closely related languages in the same domain. The authors conclude that reported parity between unsupervised and supervised methods in earlier work came from comparing full unsupervised systems against stripped-down supervised baselines.

Load-bearing premise

The paper's sweeping conclusion assumes that the one unsupervised method it tests represents the best of all fully unsupervised methods; if any untested unsupervised method beat the 500-pair baseline on a distant language pair, the central 'never outperform' claim would fall.

Editorial extensions

If this is right

  • Fair evaluation of CLWE methods must compare full component stacks; otherwise unsupervised systems appear to match supervised ones only because the baselines are deprived of self-learning and preprocessing.
  • A few hundred translation pairs (500 or 1K) are enough to make projection-based CLWE robust across all 210 language pairs, eliminating the near-zero failure mode entirely.
  • Unsupervised CLWE's main use case—distant, resource-poor, typologically diverse pairs—is where it fails most: Korean, Thai, Basque, Georgian, and Indonesian yield the largest numbers of unsuccessful setups.
  • Domain mismatch between monolingual corpora breaks unsupervised alignment, while weakly supervised methods with 500 pairs or identical-string seeds still produce useful solutions.
  • The practical contribution of unsupervised CLWE work is the robust self-learning and preprocessing toolkit (C2 and C3), which transfers to weakly supervised settings.

Reading between the lines

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

  • Editorial inference: the paper tests a single unsupervised method; running GAN-based, Gromov-Wasserstein, or other unsupervised alignment families under the same component-controlled protocol would test whether 'unsupervised always loses' is a law or a property of VECMAP.
  • Editorial inference: the component decomposition suggests a cheaper route to broad language coverage than perfecting unsupervised initialization—mining small seed dictionaries from resources like PanLex or shared numerals could be prioritized; the authors gesture at this but stop short of prescribing it.
  • Editorial inference: downstream systems that currently substitute unsupervised CLWEs into unsupervised NMT or cross-lingual IR would inherit the zero-solution failure mode; testing those end tasks on the 87 failing pairs would quantify the real-world cost.
  • Editorial inference: the near-isomorphism assumption, which the paper identifies as the root cause, could be probed by measuring how BLI score varies with a typological distance metric; if a threshold exists, it would give practitioners a cheap predictor of when unsupervised alignment will fail.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper decomposes unsupervised projection-based cross-lingual word embedding (CLWE) learning into three components: C1 (unsupervised seed dictionary extraction), C2 (iterative self-learning), and C3 (preprocessing/postprocessing). It argues that the only core difference between fully unsupervised and weakly supervised projection-based methods is C1, and that the C2/C3 machinery of VECMAP (Artetxe et al., 2018b) can be transplanted to weakly supervised baselines. Using VECMAP as the representative unsupervised method, reporting its best of three self-learning variants and averaging over five restarts, and applying the same C2/C3 components to weakly supervised baselines seeded with 500, 1K, and 5K translation pairs, the paper evaluates bilingual lexicon induction (BLI) over 210 directed pairs among 15 typologically diverse languages. It reports that the unsupervised method is near-zero on 87/210 setups, that it never beats the weakly supervised FULL+SL+SYM or FULL+SL+NOD configurations on any pair, that C3 and symmetric self-learning drive most of the improvement, and that domain mismatch between monolingual corpora is still fatal for the unsupervised method. The training and test dictionaries are released publicly.

Significance. If the result holds, it is an important negative result for fully unsupervised CLWE methods, reframing their main contribution as the self-learning and preprocessing components (C2/C3) rather than the unsupervised seed extraction (C1), and providing practical guidance to use small seed lexicons. The paper's strengths include a controlled comparison in which C2/C3 are held fixed across supervised and unsupervised conditions, a large and diverse 210-pair evaluation grid, conservative reporting for the unsupervised model (best over self-learning variants, averages over five restarts), and the release of all training and test dictionaries. The count of 87/210 near-zero setups is a concrete, falsifiable finding. The main caveat is that all unsupervised evidence comes from a single algorithm, VECMAP, so the universal negative claim in the abstract and title is broader than the experimental support.

major comments (2)
  1. [Section 2.1 and Abstract] The abstract's class-level claim that fully unsupervised CLWE methods 'still fail' and 'never surpass' weakly supervised methods is supported experimentally only by VECMAP as the unsupervised C1 component. The paper states in Section 2.1 that 'We thus focus on VECMAP in our analyses,' based on prior surveys and 'preliminary BLI tests (not shown for brevity),' but no alternative unsupervised seed-extraction method (e.g., MUSE/adversarial, Gromov-Wasserstein alignment, hyperalignment) is run under the same controlled C2/C3 conditions. Since the conclusion is a universal negative over a whole class of methods, the load-bearing assumption that VECMAP is the best possible C1 for distant pairs is plausible but unproven. I recommend either adding at least one or two alternative unsupervised methods on a subset of the grid (especially the hardest pairs, e.g., KO-* and TH-*), or explicitly scoping the paper's claims to 'the best current unsupervised method we tested' and noting that no fully unsupervised method we tested beats weak supervision.
  2. [Section 4, Table 4] The claim that FULL+SL+SYM and FULL+SL+NOD outperform the best UNSUPERVISED model 'for all 210/210 BLI setups' is central, but Table 4 does not make this pairwise comparison directly visible. Its 'Win' column counts how often each configuration is the best overall among the supervised variants, not how often it beats UNSUPERVISED. I verified the claim from the per-pair supplementary tables, but the main text should state the pairwise count explicitly or add a row such as 'FULL+SL+SYM vs UNSUPERVISED: 210-0' for each seed size. This is an auditability issue for the paper's headline result.
minor comments (4)
  1. [Supplemental Material] The text introducing the supplemental material says results are provided for dictionary sizes |D0|=1,000 and |D0|=500, but Tables 6-20 display 5k and 1k; the 500-pair results appear only for selected setups in Table 5. Please clarify which dictionary sizes are covered in the per-pair tables and where the full 500-pair results can be found.
  2. [Section 4, Figure 3 discussion] The sentence referring to 'FULL-SUPER +SYM' uses a configuration name that does not appear in Table 1; the intended configuration appears to be FULL+SL+SYM. Please correct the notation.
  3. [Section 4] The statement that dropout with self-learning is useful 'only when we work with noisy unsupervised seed lexicons' is presented as a general conclusion, but it is inferred from a specific comparison (FULL+SL vs FULL+SL+NOD) under one vocabulary size and one dropout probability. A caveat about the limited hyperparameter range would be appropriate.
  4. [Section 5] The sentence claiming that unsupervised CLWEs are 'fully substitutable with the superior weakly supervised CLWEs' in unsupervised NMT, domain adaptation, and cross-lingual IR is an extrapolation from BLI results and is not tested in this paper. It would be safer to present this as a hypothesis for future work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the BLI comparison is a controlled empirical evaluation, and the class-level generalization from VECMAP is an external-validity caveat rather than a circular reduction.

full rationale

This paper is an empirical comparison, not a derivation from assumed conclusions. The central quantitative findings—87/210 BLI setups at (near-)zero MRR for the fully unsupervised model, and FULL+SL+SYM/FULL+SL+NOD outperforming UNSUPERVISED on all 210/210 setups when C2 and C3 are held fixed—are contingent on the data and algorithms and are not baked into the experimental construction. Hyperparameters are inherited from prior work; training and test dictionaries are fixed PanLex splits; and no parameter is fitted to the target claim. In particular, the weakly supervised FULL+SL+SYM configuration is not constructed so that it must dominate the unsupervised run: if C1 produced a perfect seed dictionary, the unsupervised method could in principle match it, so the observed superiority is falsifiable. The only in-scope issue is the breadth of the class-level conclusion: the paper tests "fully unsupervised" only through VECMAP (Section 2.1: "We thus focus on VECMAP in our analyses"), and the choice of representative is justified by prior surveys including the authors' own Glavaš et al. (2019) plus "our own preliminary BLI tests (not shown for brevity)". However, the same sentence cites the independent evaluation of Doval et al. (2019), and the prior survey is itself an empirical, externally checkable comparison rather than an unverified self-citation. The gap between what was tested (VECMAP) and what is claimed ("fully unsupervised CLWE approaches... never surpass") is a generalizability or reporting caveat, not a circular reduction. None of the enumerated circularity patterns is exhibited.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

This is an evaluation paper, so the ledger contains no newly invented entities or fitted model parameters. The central claim depends on the chosen evaluation design, the representativeness of the single unsupervised method, and the reliability of automatically constructed dictionaries.

free parameters (2)
  • Seed dictionary sizes for weakly supervised methods = 500, 1000, 5000 pairs
    The comparison is anchored to these sizes. For smaller seeds (e.g., 200), the supervised method still wins per Figure 3, but the paper does not map the threshold where unsupervised could become competitive.
  • Unsupervised hyperparameters (top-k vocab sizes, dropout, CSLS) = seed: 4K words; self-learning: 20K words; dropout p=0.1; CSLS k=10
    Adopted from Artetxe et al. (2018b) and fixed. A hyperparameter sweep could change the failure rate for the unsupervised method, so the headline numbers are conditional on these hand-chosen values.
assumptions (4)
  • domain assumption Bilingual lexicon induction with MRR is a valid proxy for cross-lingual embedding quality.
    The main study evaluates only BLI, citing Glavaš et al. (2019) for correlation with retrieval and NLI tasks; if BLI is not representative for downstream tasks, conclusions may not transfer.
  • domain assumption VECMAP is representative of the most robust fully unsupervised CLWE methods.
    The paper selects VECMAP based on comparative studies (Glavaš et al. 2019; Doval et al. 2019) and preliminary tests, rather than evaluating all state-of-the-art unsupervised methods in the same framework.
  • domain assumption PanLex-derived translation pairs are accurate enough for training and test dictionaries.
    Training/test dictionaries are automatically extracted from PanLex using a predefined confidence threshold; no manual verification or noise analysis is reported.
  • domain assumption Projection-based CLWE methods require approximate isomorphism between monolingual embedding spaces.
    The paper uses the violation of isomorphism to explain unsupervised failures for distant pairs; this assumption is discussed, not directly measured or verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Do We Really Need Fully Unsupervised Cross-Lingual Embeddings?." pith.science (2026). https://pith.science/paper/GWQLFYYI

@misc{pith2026190901638,
  author       = {Pith},
  title        = {Pith review of: Do We Really Need Fully Unsupervised Cross-Lingual Embeddings?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GWQLFYYI}},
  note         = {Machine review of arXiv:1909.01638}
}
read the original abstract

Recent efforts in cross-lingual word embedding (CLWE) learning have predominantly focused on fully unsupervised approaches that project monolingual embeddings into a shared cross-lingual space without any cross-lingual signal. The lack of any supervision makes such approaches conceptually attractive. Yet, their only core difference from (weakly) supervised projection-based CLWE methods is in the way they obtain a seed dictionary used to initialize an iterative self-learning procedure. The fully unsupervised methods have arguably become more robust, and their primary use case is CLWE induction for pairs of resource-poor and distant languages. In this paper, we question the ability of even the most robust unsupervised CLWE approaches to induce meaningful CLWEs in these more challenging settings. A series of bilingual lexicon induction (BLI) experiments with 15 diverse languages (210 language pairs) show that fully unsupervised CLWE methods still fail for a large number of language pairs (e.g., they yield zero BLI performance for 87/210 pairs). Even when they succeed, they never surpass the performance of weakly supervised methods (seeded with 500-1,000 translation pairs) using the same self-learning procedure in any BLI setup, and the gaps are often substantial. These findings call for revisiting the main motivations behind fully unsupervised CLWE methods.

Figures

Figures reproduced from arXiv: 1909.01638 by the authors.

Figure 1
Figure 1. General unsupervised CLWE approach. algorithms for unsupervised seed dictionary induc￾tion differ, they all strongly rely on the assump￾tion of similar topological structure between the two pretrained monolingual spaces. Once the seed dictionary is obtained, the two-step iterative self￾learning procedure (C2) takes place: 1) a dictio￾nary D(k) is first used to learn the joint space Y (k) = XW(k) x ∪ ZW(k) z ; 2) the… view at source ↗
Figure 2
Figure 2. A comparison of average BLI scores with different seed dictionary sizes D0 between a fully un￾supervised method (UNSUPER), a supervised method without self-learning (SUPER), and two best perform￾ing weakly supervised methods with self learning (+SL+NOD and +SL+SYM). While SUPER without self￾learning displays a steep drop in performance with smaller seed dictionaries, there is only a slight decrease when self-learnin… view at source ↗
Figure 3
Figure 3. A comparison of BLI scores on “easy” (i.e., similar) language pairs between the fully [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: BLI scores with the (most robust) fully UNSUPERVISED model for different language pairs when mono￾lingual word embeddings are trained on dissimilar domains: parliamentary proceedings (EuroParl), Wikipedia (Wiki), and medical corpora (EMEA). Training and test data are t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 53 canonical work pages

  1. [1]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

  2. [2]

    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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Oliver Adams, Adam Makarucha, Graham Neubig, Steven Bird, and Trevor Cohn. 2017. http://www.aclweb.org/anthology/E17-1088 Cross-lingual word embeddings for low-resource language modeling . In Proceedings of EACL, pages 937--947

    work page 2017

  4. [4]

    Rami Al-Rfou, Bryan Perozzi, and Steven Skiena. 2013. http://www.aclweb.org/anthology/W13-3520 Polyglot: D istributed word representations for multilingual NLP . In Proceedings of CoNLL, pages 183--192

    work page 2013

  5. [5]

    Jean Alaux, Edouard Grave, Marco Cuturi, and Armand Joulin. 2019. http://arxiv.org/abs/1811.01124 Unsupervised hyperalignment for multilingual word embeddings . In Proceedings of ICLR

    arXiv 2019

  6. [6]

    David Alvarez-Melis and Tommi Jaakkola. 2018. http://aclweb.org/anthology/D18-1214 Gromov-W asserstein alignment of word embedding spaces . In Proceedings of EMNLP, pages 1881--1890

    work page 2018

  7. [7]

    Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2017. http://aclweb.org/anthology/P17-1042 Learning bilingual word embeddings with (almost) no bilingual data . In Proceedings of ACL, pages 451--462

    work page 2017

  8. [8]

    Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2018 a . https://www.aaai.org/ocs/index.php/AAAI/AAAI18/paper/view/16935 Generalizing and improving bilingual word embedding mappings with a multi-step framework of linear transformations . In Proceedings of AAAI, pages 5012--5019

    work page 2018

Show all 63 references
  1. [9]

    Mikel Artetxe, Gorka Labaka, and Eneko Agirre. 2018 b . http://aclweb.org/anthology/P18-1073 A robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings . In Proceedings of ACL, pages 789--798

    2018

  2. [10]

    Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. 2018 c . http://arxiv.org/abs/1710.11041 Unsupervised neural machine translation . In Proceedings of ICLR

    2018 arXiv

  3. [11]

    Timothy Baldwin, Jonathan Pool, and Susan Colowick. 2010. http://www.Association for Computational Linguisticsweb.org/anthology/C10-3010 PanLex and LEXTRACT: T ranslating all words of all languages of the world . In Proceedings of COLING (Demo Papers), pages 37--40

    2010

  4. [12]

    Antonio Valerio Miceli Barone. 2016. http://arxiv.org/pdf/1608.02996.pdf Towards cross-lingual distributed representations without parallel text trained with adversarial autoencoders . In Proceedings of the 1st Workshop on Representation Learning for NLP, pages 121--126

    2016 arXiv

  5. [13]

    Anthony Bell and Terrence Sejnowski. 1997. http://arxiv.org/abs/9809069v1 The 'Independent Components' of Natural Scenes are Edge Filters . Vision Research

    1997

  6. [14]

    Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. http://arxiv.org/abs/1607.04606 Enriching word vectors with subword information . Transactions of the ACL, 5:135--146

    2017 arXiv

  7. [15]

    Danqi Chen and Christopher Manning. 2014. https://www.aclweb.org/anthology/D14-1082 A fast and accurate dependency parser using neural networks . In Proceedings of EMNLP, pages 740--750

    2014

  8. [16]

    Xilun Chen and Claire Cardie. 2018. http://aclweb.org/anthology/D18-1024 Unsupervised multilingual word embeddings . In Proceedings of EMNLP, pages 261--270

    2018

  9. [17]

    Alexis Conneau, Guillaume Lample, Marc'Aurelio Ranzato, Ludovic Denoyer, and Herv \'e J \'e gou. 2018 a . https://arxiv.org/abs/1710.04087 Word translation without parallel data . In Proceedings of ICLR

    2018 arXiv

  10. [18]

    Alexis Conneau, Ruty Rinott, Guillaume Lample, Adina Williams, Samuel Bowman, Holger Schwenk, and Veselin Stoyanov. 2018 b . http://aclweb.org/anthology/D18-1269 XNLI: E valuating cross-lingual sentence representations . In Proceedings of EMNLP, pages 2475--2485

    2018

  11. [19]

    Mielke, Jason Eisner, and Brian Roark

    Ryan Cotterell, Sebastian J. Mielke, Jason Eisner, and Brian Roark. 2018. https://ryancotterell.github.io/papers/cotterell+alc.naacl18.pdf Are all languages equally hard to language-model? In Proceedings of NAACL-HLT

    2018

  12. [20]

    Zi-Yi Dou, Zhi-Hao Zhou, and Shujian Huang. 2018. http://aclweb.org/anthology/D18-1062 Unsupervised bilingual lexicon induction via latent variable models . In Proceedings of EMNLP, pages 621--626

    2018

  13. [21]

    Yerai Doval, Jose Camacho-Collados, Luis Espinosa-Anke, and Steven Schockaert. 2019. https://arxiv.org/pdf/1908.07742.pdf On the robustness of unsupervised and semi-supervised cross-lingual word embedding learning . CoRR, abs/1908.07742

    2019 arXiv

  14. [22]

    Long Duong, Hiroshi Kanayama, Tengfei Ma, Steven Bird, and Trevor Cohn. 2016. http://aclweb.org/anthology/D16-1136 Learning crosslingual word embeddings without bilingual corpora . In Proceedings of EMNLP, pages 1285--1295

    2016

  15. [23]

    Manaal Faruqui and Chris Dyer. 2014. http://aclweb.org/anthology/E14-1049 Improving vector space word representations using multilingual correlation . In Proceedings of EACL, pages 462--471

    2014

  16. [24]

    Daniela Gerz, Ivan Vuli \'c , Edoardo Maria Ponti, Roi Reichart, and Anna Korhonen. 2018. https://www.aclweb.org/anthology/D18-1029 On the relation between linguistic typology and (limitations of) multilingual language modeling . In Proceedings of EMNLP, pages 316--327

    2018

  17. [25]

    Goran Glava s , Marc Franco-Salvador, Simone P Ponzetto, and Paolo Rosso. 2018. https://arxiv.org/abs/1801.06436 A resource-light method for cross-lingual semantic textual similarity . Knowledge-Based Systems, 143:1--9

    2018 arXiv

  18. [26]

    Goran Glava s , Robert Litschko, Sebastian Ruder, and Ivan Vuli \'c . 2019. https://www.aclweb.org/anthology/P19-1070 How to (properly) evaluate cross-lingual word embeddings: O n strong baselines, comparative analyses, and some misconceptions . In Proceedings of ACL, pages 710--721

    2019

  19. [27]

    Stephan Gouws, Yoshua Bengio, and Greg Corrado. 2015. http://jmlr.org/proceedings/papers/v37/gouws15.html BilBOWA: F ast bilingual distributed representations without word alignments . In Proceedings of ICML, pages 748--756

    2015

  20. [28]

    Stephan Gouws and Anders S gaard. 2015. https://www.aclweb.org/anthology/N15-1157 Simple task-specific bilingual word embeddings . In Proceedings of NAACL-HLT, pages 1386--1390

    2015

  21. [29]

    Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. 2018. http://aclweb.org/anthology/L18-1550 Learning word vectors for 157 languages . In Proceedings of LREC, pages 3483--3487

    2018

  22. [30]

    Jiang Guo, Wanxiang Che, David Yarowsky, Haifeng Wang, and Ting Liu. 2015. http://aclweb.org/anthology/P15-1119 Cross-lingual dependency parsing based on distributed representations . In Proceedings of ACL, pages 1234--1244

    2015

  23. [31]

    Geert Heyman, Bregt Verreet, Ivan Vuli \'c , and Marie-Francine Moens. 2019. https://www.aclweb.org/anthology/N19-1188 Learning unsupervised multilingual word embeddings with incremental multilingual hubs . In Proceedings of NAACL-HLT, pages 1890--1902

    2019

  24. [32]

    Geert Heyman, Ivan Vuli \' c , and Marie-Francine Moens. 2017. http://aclweb.org/anthology/E17-1102 Bilingual lexicon induction by learning to combine word-level and character-level representations . In Proceedings of EACL, pages 1085--1095

    2017

  25. [33]

    Yedid Hoshen and Lior Wolf. 2018. http://aclweb.org/anthology/D18-1043 Non-adversarial unsupervised word translation . In Proceedings of EMNLP, pages 469--478

    2018

  26. [34]

    Colowick

    David Kamholz, Jonathan Pool, and Susan M. Colowick. 2014. http://www.lrec-conf.org/proceedings/lrec2014/summaries/1029.html Panlex: B uilding a resource for panlingual lexical translation . In Proceedings of LREC, pages 3145--3150

    2014

  27. [35]

    Yova Kementchedjhieva, Sebastian Ruder, Ryan Cotterell, and Anders S gaard. 2018. https://www.aclweb.org/anthology/K18-1021 Generalizing P rocrustes analysis for better bilingual dictionary induction . In Proceedings of CoNLL, pages 211--220

    2018

  28. [36]

    Yunsu Kim, Jiahui Geng, and Hermann Ney. 2018. http://aclweb.org/anthology/D18-1101 Improving unsupervised word-by-word translation with language model and denoising autoencoder . In Proceedings of EMNLP, pages 862--868

    2018

  29. [37]

    Alexandre Klementiev, Ivan Titov, and Binod Bhattarai. 2012. http://aclweb.org/anthology/C12-1089 Inducing crosslingual distributed representations of words . Proceedings of COLING, pages 1459--1474

    2012

  30. [38]

    Philipp Koehn. 2005. http://www.statmt.org/europarl/ Europarl: A parallel corpus for statistical machine translation . In Proceedings of the 10th Machine Translation Summit (MT SUMMIT) , pages 79--86

    2005

  31. [39]

    Guillaume Lample, Alexis Conneau, Ludovic Denoyer, and Marc'Aurelio Ranzato. 2018 a . https://openreview.net/forum?id=rkYTTf-AZ Unsupervised machine translation using monolingual corpora only . In Proceedings of ICLR

    2018

  32. [40]

    Guillaume Lample, Myle Ott, Alexis Conneau, Ludovic Denoyer, and Marc'Aurelio Ranzato. 2018 b . http://aclweb.org/anthology/D18-1549 Phrase-based & neural unsupervised machine translation . In Proceedings of EMNLP, pages 5039--5049

    2018

  33. [41]

    Robert Litschko, Goran Glava s , Ivan Vulic, and Laura Dietz. 2019. https://dl.acm.org/citation.cfm?doid=3331184.3331324 Evaluating resource-lean cross-lingual embedding models in unsupervised retrieval . In Proceedings of SIGIR, pages 1109--1112

    2019

  34. [42]

    Robert Litschko, Goran Glava s , Simone Paolo Ponzetto, and Ivan Vuli\' c . 2018. https://arxiv.org/abs/1805.00879 Unsupervised cross-lingual information retrieval using monolingual data only . In Proceedings of SIGIR, pages 1253--1256

    2018 arXiv

  35. [43]

    Noa Yehezkel Lubin, Jacob Goldberger, and Yoav Goldberg. 2019. https://www.aclweb.org/anthology/N19-1045 Aligning vector-spaces with noisy supervised lexicon . In Proceedings of NAACL-HLT, pages 460--465

    2019

  36. [44]

    Tomas Mikolov, Quoc V Le, and Ilya Sutskever. 2013 a . https://arxiv.org/abs/1309.4168 Exploiting similarities among languages for machine translation . CoRR, abs/1309.4168

    2013 arXiv

  37. [45]

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Gregory S Corrado, and Jeffrey Dean. 2013 b . https://papers.nips.cc/paper/5021-distributed-representations-of-words-and-phrases-and-their-compositionality.pdf Distributed Representations of Words and Phrases and their Compositionality ...

    2013

  38. [46]

    Tasnim Mohiuddin and Shafiq Joty. 2019. https://www.aclweb.org/anthology/N19-1386 Revisiting adversarial autoencoder for unsupervised word translation with cycle consistency and improved training . In Proceedings of NAACL-HLT, pages 3857--3867

    2019

  39. [47]

    Ndapa Nakashole. 2018. http://aclweb.org/anthology/D18-1047 NORMA: N eighborhood sensitive maps for multilingual word embeddings . In Proceedings of EMNLP, pages 512--522

    2018

  40. [48]

    Aitor Ormazabal, Mikel Artetxe, Gorka Labaka, Aitor Soroa, and Eneko Agirre. 2019. https://www.aclweb.org/anthology/P19-1492 Analyzing the limitations of cross-lingual word embedding mappings . In Proceedings of ACL, pages 4990--4995

    2019

  41. [49]

    Sebastian Ruder, Ivan Vuli\' c , and Anders S gaard. 2019. https://doi.org/10.1613/jair.1.11640 A survey of cross-lingual word embedding models . Journal of Artificial Intelligence Research, 65:569--631

    2019 doi

  42. [50]

    Smith, David H.P

    Samuel L. Smith, David H.P. Turban, Steven Hamblin, and Nils Y. Hammerla. 2017. https://arxiv.org/abs/1702.03859 Offline bilingual word vectors, orthogonal transformations and the inverted softmax . In Proceedings of ICLR

    2017 arXiv

  43. [51]

    Anders S gaard, Z eljko Agi \' c , H \'e ctor Mart \'i nez Alonso, Barbara Plank, Bernd Bohnet, and Anders Johannsen. 2015. http://aclweb.org/anthology/P15-1165 Inverted indexing for cross-lingual NLP . In Proceedings of ACL, pages 1713--1722

    2015

  44. [52]

    Anders S gaard, Sebastian Ruder, and Ivan Vuli \' c . 2018. http://aclweb.org/anthology/P18-1072 On the limitations of unsupervised bilingual dictionary induction . In Proceedings of ACL, pages 778--788

    2018

  45. [53]

    J\"org Tiedemann. 2009. http://stp.lingfil.uu.se/ joerg/published/ranlp-V.pdf News from OPUS - A collection of multilingual parallel corpora with tools and interfaces . In Proceedings of RANLP, pages 237--248

    2009

  46. [54]

    Joseph Turian, Lev-Arie Ratinov, and Yoshua Bengio. 2010. https://www.aclweb.org/anthology/P10-1040 Word representations: A simple and general method for semi-supervised learning . In Proceedings of ACL, pages 384--394

    2010

  47. [55]

    Ivan Vuli \' c and Anna Korhonen. 2016. http://aclweb.org/anthology/P16-1024 On the role of seed lexicons in learning bilingual word embeddings . In Proceedings of ACL, pages 247--257

    2016

  48. [56]

    Ivan Vuli \' c and Marie-Francine Moens. 2015. https://dl.acm.org/citation.cfm?id=2767752 Monolingual and cross-lingual information retrieval models based on (bilingual) word embeddings . In Proceedings of SIGIR, pages 363--372

    2015

  49. [57]

    Ivan Vuli\' c , Nikola Mrk s i\' c , and Anna Korhonen. 2017. https://www.aclweb.org/anthology/D17-1269 Cross-lingual induction and transfer of verb classes based on word vector space specialisation . In Proceedings of EMNLP, pages 2536--2548

    2017

  50. [58]

    S ren Wichmann, Andr \'e M \"u ller, Viveka Velupillai, Cecil H Brown, Eric W Holman, Pamela Brown, Sebastian Sauppe, Oleg Belyaev, Matthias Urban, Zarina Molochieva, et al. 2018. https://asjp.clld.org/ The ASJP database (version 18)

    2018

  51. [59]

    Chao Xing, Dong Wang, Chao Liu, and Yiye Lin. 2015. http://aclweb.org/anthology/N15-1104 Normalized word embedding and orthogonal transform for bilingual word translation . In Proceedings of NAACL-HLT, pages 1006--1011

    2015

  52. [60]

    Mozhi Zhang, Keyulu Xu, Ken-ichi Kawarabayashi, Stefanie Jegelka, and Jordan Boyd-Graber. 2019. https://www.aclweb.org/anthology/P19-1307 Are girls neko or sh \=o jo? C ross-lingual alignment of non-isomorphic embeddings with iterative normalization . In Proceedings of ACL, pa...

    2019

  53. [61]

    Yuan Zhang, David Gaddy, Regina Barzilay, and Tommi Jaakkola. 2016. http://aclweb.org/anthology/N16-1156 Ten pairs to tag -- Multilingual POS tagging via coarse mapping between embeddings . In Proceedings of NAACL-HLT, pages 1307--1317

    2016

  54. [62]

    George Kingsley Zipf. 1949. https://onlinelibrary.wiley.com/doi/abs/10.1002/1097-4679 Human behavior and the principle of least effort: A n introduction to human ecology

    1949 doi

  55. [63]

    Yftah Ziser and Roi Reichart. 2018. https://www.aclweb.org/anthology/D18-1022 Deep pivot-based modeling for cross-language cross-domain transfer with minimal guidance . In Proceedings of EMNLP, pages 238--249

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.