Skip to content

Spanish glossary

This page codifies the terminology rules from LYNX_CORTEX_ADDENDUM.md §A17. Every Spanish translation in the repo (*.es.md files) follows these rules.

🇪🇸 Esta página define los términos a usar al traducir el currículo al español. Cada archivo *.es.md del repositorio sigue estas reglas.

1. English loanwords (keep verbatim, no parenthesis)

These English words are in active Spanish technical use; translating them is unnatural. Keep them as-is. No parenthetical gloss.

Term Notes
switch network device — everyone says switch
router network device — enrutador exists but router is dominant
framework marco exists but is rarely used
software, hardware, firmware universal
bug, commit, push, pull request, merge, rebase, branch git workflow vocabulary
endpoint, payload, request, response HTTP / API vocabulary
cache, pipeline, batch, mini-batch systems / ML vocabulary
script, log, query universal
token, tokenizer, tokenization NLP-specific
embedding, attention, transformer ML architectures
prompt, prompt injection, jailbreak LLM vocabulary
dataset, feature, label data science
notebook Jupyter / Marimo
repo, repository git — repositorio works too; either is fine
default, string, boolean, array universal
wrapper, helper, mock, stub engineering
release, rollback, rollout, deploy, deployment ops
bandwidth or ancho de banda both acceptable — see §2 for the preferred form

2. Spanish term first, English in parentheses (on first mention per page)

These have established Spanish translations and should be translated, with the English term in parentheses on the first mention of a page:

Spanish term English original
Tecnologías de la Información (IT) Information Technology
aprendizaje automático (machine learning) machine learning
aprendizaje profundo (deep learning) deep learning
aprendizaje por refuerzo (reinforcement learning) reinforcement learning
red neuronal neural network
capa layer (e.g. capa de atención)
pérdida / función de pérdida loss / loss function
descenso de gradiente gradient descent
retropropagación (backpropagation) backpropagation
regularización regularization
sobreajuste (overfitting) overfitting
infraajuste (underfitting) underfitting
ajuste fino (fine-tuning) fine-tuning
sintonización tuning (general)
inferencia inference
entrenamiento training
muestra / muestreo sample / sampling
seguridad security (vs. seguro = safety)
amenaza threat
vulnerabilidad vulnerability
ingeniería de prompts prompt engineering
modelo de lenguaje language model
modelo de lenguaje grande (LLM) large language model
ventana de contexto context window
memoria caché cache memory (system memory tier; the cache term in §1 covers software caches)
jerarquía de memoria memory hierarchy
ancho de banda bandwidth
latencia latency
rendimiento throughput / performance
confiabilidad reliability
disponibilidad availability
portabilidad portability
reproducibilidad reproducibility
inmutable immutable
flujo de trabajo workflow
revisión por pares peer review
informe / reporte report
punto de control checkpoint
semilla seed (as in seed_everything)

3. Acronyms

Use Spanish form first if an established Spanish acronym exists; otherwise stay English. Always gloss on first mention per page.

Spanish English Notes
IA AI Inteligencia Artificial
PLN NLP Procesamiento del Lenguaje Natural
MCO OLS Mínimos Cuadrados Ordinarios
(none) CPU, GPU, RAM, ROM, SSD hardware acronyms — stay English
(none) HTTP, HTTPS, TCP, IP, DNS protocol acronyms — stay English
(none) JSON, YAML, XML, SQL, CSV format acronyms — stay English
(none) API, SDK, CLI, IDE tooling acronyms — stay English
(none) BPE, MLP, MoE, RAG, RLHF, DPO, LoRA, QLoRA, GPT, ViT, CLIP, KV cache ML acronyms — stay English
(none) DoD, SLO, SLA, KPI ops acronyms — stay English
(none) CI, CD, PR git/CI acronyms — stay English

4. Identifier / path / command rules

These are never translated, regardless of context:

  • File and directory paths: learners/borja/profile.md, src/miniportal/, docs/phase-00-foundations/theory/00-motivation.md.
  • Shell commands and recipes: just setup, uv sync --frozen, git clone, ruff check.
  • Code identifiers in inline code spans: seed_everything(seed), manifest.json, PHASE_00_REPORT.md.
  • The 20 verbs of the §A13 corpus: work, play, walk, talk, listen, watch, study, finish, start, look, want, like, be, have, do, go, come, see, eat, write — these are the model's input vocabulary.
  • Spec section refs: §A13, §7.2 of LYNX_CORTEX.md.
  • Conventional Commits prefixes: feat:, fix:, docs:, phase:, chore:.

5. Tone and regional register

Spanish translations target Spain-Spanish technical register (the active learner is in Spain). Where Latin-American variants differ:

  • ordenador (not computadora)
  • vídeo (not video)
  • impersonal / forms in instructional text, usted only when formal address is unavoidable

When a sentence reads more naturally with the English term inline, defer to the English term (per §1). Forced translations like enrutador over router or marco over framework are awkward and not used.

6. Examples

English source:

The router forwards packets; the switch keeps a MAC address table per port.

Spanish mirror:

El router reenvía paquetes; el switch mantiene una tabla de direcciones MAC por puerto.


English source:

Backpropagation is gradient descent applied to a computation graph using the chain rule.

Spanish mirror:

La retropropagación (backpropagation) es descenso de gradiente aplicado a un grafo de cómputo usando la regla de la cadena.


English source:

Run just setup to provision the environment; this calls uv sync --extra core --extra dev.

Spanish mirror:

Ejecuta just setup para aprovisionar el entorno; esto invoca uv sync --extra core --extra dev.


English source:

Phase 32 ships a grammar tutor agent: read an English sentence, propose conjugation corrections.

Spanish mirror:

La Fase 32 entrega un agente tutor de gramática: lee una frase en inglés y propone correcciones de conjugación.

7. When in doubt

If a term is not in §1–§3, prefer the form most natural for a Spanish-speaking software engineer reading their own native technical literature. When unsure, leave the English term and flag the case to the maintainer — adding a new row to this glossary is the canonical fix, not picking an ad-hoc translation.