pylazaro: a Python package for anglicism extraction in Spanish
cs.CL
Quick verdict
pylazaro unifies five Spanish anglicism-extraction models (F1 up to 0.86, well above LLMs) in one solid package; useful tool paper, but no new experiments.
Summary
This tool paper introduces pylazaro, an open-source Python package (MIT license, PyPI, readthedocs, live HuggingFace demo) that extracts unassimilated lexical borrowings — mostly anglicisms — from Spanish text. The package wraps a unified Lazaro interface around five sequence-labeling models from the author's prior work (Alvarez-Mellado and Lignos, 2022): a BiLSTM-CRF with code-switching embeddings (default; F1 0.86), a BiLSTM-CRF with BETO/BERT embeddings (0.84), mBERT (0.84), BETO (0.83), and a handcrafted-feature CRF. Models tag tokens with BIO encoding, distinguish ENG vs. OTHER labels, and return tuples, dictionaries, spans, or per-token tags; switching models requires changing one line. Against general-purpose LLMs evaluated in other studies (González et al., 2026; Alvarez-Mellado, 2025), the fine-tuned models (109–226M parameters) score 0.83–0.86 F1 versus 0.39 for Llama3-8B and at most 0.32 among 23 LLMs, and run on a CPU-only laptop. Reported adoption includes 58,000+ PyPI downloads, 63,000+ HuggingFace model downloads, and its role powering Observatorio Lázaro, which has collected over 2 million anglicisms since 2020. Main limitations: models were trained on European Spanish newspaper text; performance on other domains and varieties is unevaluated; only unassimilated borrowings are detected; non-English borrowings are underrepresented; and the paper itself runs no new experiments, reusing numbers from prior publications.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- pylazaro is, to the authors' knowledge, the first publicly available library for identifying unassimilated borrowings in running Spanish text.
- Its five fine-tuned models (F1 0.83–0.86) substantially outperform general-purpose LLMs (best ≤0.39) on anglicism extraction while being one to two orders of magnitude smaller (109–226M vs 9–40B parameters).
- A single unified interface exposes models built on different libraries (flair, Transformers, sklearn-crfsuite), switchable with a one-line code change.
- The library has achieved tangible adoption: 58,000+ PyPI downloads and it powers Observatorio Lázaro, which has collected over 2 million anglicisms since 2020.
Strengths
Fills a genuine tooling gap
Sections 1–2 argue no standard NLP library (spaCy, stanza) offers borrowing detection and prior ADoBo models were unreleased research code; pylazaro packages them installably (PyPI, MIT, readthedocs, HF demo).
Clean unified architecture over heterogeneous backends
Section 3.4: the Lazaro class dispatches to per-library classifiers (FlairClassifier, TransformersClassifier, CRFClassifier), fuses and aligns BIO subtokens, and wraps results in one output object; switching models is a one-line change (Listing 2) and new models need only a new classifier.
Large, quantified margin over LLMs at tiny scale
Demonstrated real-world adoption
Candid limitations section
Weaknesses
No new experiments
Cross-study LLM comparison comparability is unverified
Adoption metrics are noisy proxies
CPU-runnable claim lacks measurement
Domain gap acknowledged but unquantified
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a tool/resource paper: the deliverable is pylazaro, an MIT-licensed Python package (PyPI, readthedocs, HuggingFace Spaces demo) that extracts unassimilated lexical borrowings — mostly anglicisms — from Spanish text. The authors stake four claims (Section 1): pylazaro is the first publicly available library for detecting borrowings in running text; it offers a unified interface to five sequence-labeling models built on different libraries; these fine-tuned models decisively beat general-purpose LLMs (F1 0.83–0.86 vs ≤0.39, Table 1); and the library has real adoption (58,000+ PyPI downloads; powers Observatorio Lázaro). Critically, the comparative claim rests entirely on numbers imported from prior publications — the paper trains nothing and runs no new evaluation.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is pylazaro and what does it do?
pylazaro is an open-source Python package that automatically finds unassimilated lexical borrowings — mostly anglicisms, i.e., recent English loanwords — in Spanish text. It runs five sequence-labeling models behind one interface, is installable from PyPI, documented on readthedocs, and available as a no-install demo on HuggingFace Spaces.
How accurate is pylazaro compared to using a general-purpose LLM?
On the standard newspaper evaluation set, the best pylazaro model (a BiLSTM-CRF) scores about 0.86 F1, while general-purpose LLMs tested in prior studies scored 0.39 or lower — the best of 23 LLMs reached only 0.32. The fine-tuned models are also far smaller (109–226M parameters) and run on a laptop without a GPU.
What are pylazaro's main limitations?
The models were trained on European Spanish newspaper text, so accuracy on social media or Latin American Spanish is not established; only unassimilated borrowings are detected (orthographically adapted loanwords are missed); non-English borrowings are underrepresented; and the CRF model requires extra installation dependencies.
Is pylazaro free to use?
Yes. It is MIT-licensed and free on PyPI, its models are freely downloadable from the HuggingFace Hub, and there is a free live demo on HuggingFace Spaces with no installation required.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1–2 (architecture diagram, demo screenshot) cannot be inspected from the extracted text; the architecture is judged from Section 3.4 prose and captions.
- The shared 'evaluation set' behind Table 1 is never named or sized in the text; it is assumed to be the ADoBo evaluation data from the cited prior work.
- Extraction appears complete (Sections 1–7, Limitations, and full references are present); no visible truncation.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.