crossrs 0.2.1

Command-line tool for learning foreign languages through reverse translation of word-based sentences

CrossRS

CrossRS is a command-line tool for improving language production skills through reverse translation exercises. Given a corpus in your target language, CrossRS translates sentences into a source language you already know and asks you to translate them back, reinforcing vocabulary and grammar through word-based spaced repetition. Because CrossRS uses GPT under the hood, you must have a paid OpenAI account and an API key to run it.

How It Works

CrossRS focuses on words sorted by their frequency in the corpus. You learn the most common ones first. Each study round:

1. CrossRS picks a sentence containing the next word to learn.
2. The sentence is translated into your source language and shown to you.
3. You translate it back into the target language.
4. CrossRS evaluates your translation and provides feedback — either a ✅ confirmation or a ❌ with a highlighted diff showing the minimal corrections needed.

Sentences you translate correctly on the first try are scheduled for a single review in 29 days 20 hours. Otherwise, they enter a spaced-repetition queue with reviews at 20 hours, 6 days 20 hours, and 29 days 20 hours before being marked as learned.

Installation

Install Python 3.13 or later and pipx, then run:

pipx install crossrs       # install
pipx upgrade crossrs       # upgrade
pipx uninstall crossrs     # uninstall

Initialize a New Language

Prepare a plain-text file that contains one sentence per line in the language you want to learn. For example, you can download a monolingual corpus from OPUS. Then run:

crossrs init <target-lang> <corpus>

<target-lang> is a language code (e.g., de, fr, uk), and <corpus> is the path to the corpus file.

Study a Language

crossrs study <target-lang> <source-lang> [--threshold T] [--model <GPT_MODEL>] [--api-key <OPENAI_KEY>]

• <target-lang> — the language code you initialized earlier.
• <source-lang> — the language you want sentences translated into (e.g., en).
• --threshold / -t — the learnedness threshold for words (default: 3). A word is considered fully learned once it has appeared in this many learned sentences.
• --model — the GPT model to use for translation and evaluation.
• --api-key — your OpenAI API key.

Instead of passing --model and --api-key each time, you can set the environment variables CROSSRS_MODEL and CROSSRS_API_KEY.

View Your Progress

crossrs stats <target-lang> [--threshold T]

Displays:

• Sentences: learned + in queue / total
• Words: learned / total, with word-level coverage
• Total rounds: the number of translation attempts so far

Other Commands

crossrs path <target-lang>               # show the path to the language data file
crossrs delete <target-lang> [--force]   # delete the language data file; use --force to skip the confirmation prompt