TL;DR: With the guidance of my mentor, Jonathan, I gathered a list of Swahili words from Hunspell. I then used our tools to add clear definitions from Wiktionary only for the words on that list. I also created a new workflow that relies on configuration files for each list or language. I separated paths into command line flags to avoid hard-wiring the tool to specific project folders. Finally, we ensured the output remains consistent with the current Wordnik-style importer. Now, anyone can add a new language or word list by creating a simple config file and running a few commands.

Why Swahili, and why now?

Swahili is widely spoken in East Africa yet often underrepresented in software. Crosswords are a fun way to learn vocabulary and celebrate language: so adding Swahili to GNOME Crosswords felt both useful and exciting.

alt text

Data sources at a glance

  • Word list: Hunspell Swahili dictionary (list for what words to include).

  • Definitions: Wiktionary (via our raw wiktextract JSONL dump).

  • Format goal: Output that stays compatible with the existing Broda‑style importer & GNOME word‑list tooling.

The tooling

  • def-extractor.py (in tools/wiktionary-extractor/): filters Wiktionary entries and produces a clean, structured list with parts of speech, tags, and definitions.

  • Meson/Ninja integration: convenient targets like build-wordlist-defs for repeatable builds.

alt text

Initially, the extractor logic assumed certain directory layouts and enums. To support Swahili and future languages cleanly, Jonathan and I focused on making it configurable and separating paths from the code.

What we changed and why

  • 1) Config-first workflow (new .conf files)

We introduced per-language config files so each word list can declare:

Input sources (Hunspell list, Wiktionary dump path)

Filters (minimum length, allowed parts of speech, tags, etc.)

Output targets (where the filtered .jsonl/artifacts go)

This makes adding a new language a configuration task instead of a code change.

Here’s the actual config we wrote for Swahili:

[Word List]
DisplayName=Swahili
Identifier=swahili
Source=sw_KE.dic
Importer=wordnik
Locale=sw
Definitions=True
Visibility=editor
MinLength=2
MaxLength=21
Threshold=50
Alphabet=ABCDEFGHIJKLMNOPQRSTUVWXYZ

2) Decoupled paths via CLI flags

The extractor no longer depends on hard-coded project paths. Important information can be passed from the command line. This change makes Meson rules and one-off local runs much easier.

Example (paths are illustrative):

python3 tools/wiktionary-extractor/def-extractor.py \
        --config word-lists/swahili/swahili.conf \
        --dictionary raw-wiktextract-data.jsonl \
        --word-lists-dir wordlist-tools/swahili \
        --output-dir swahili \

alt text

Swahili: end‑to‑end workflow

Here’s the exact flow we used for Swahili:

alt text

  • Dropped in the Hunspell list \ Placed the Swahili .dic under word-lists/swahili/.

  • Pointed the config at our inputs \ Created configs/swahili.conf with the inputs, filters, and outputs you want.

alt text

Build & test

Add your word list to the meson.build

    1. Open the file: 
      word-list-tools/word-lists/meson.build
      1. Locate the section that defines the word lists:

# List of word lists to build
word_lists = ['broda',
              'player',
              'test',
              'wordnik',
              'swahili']

Add your word list to this array (for example, ‘swahili’ above).

Once your config file is added, run the following from the project root to generate the GResource files.

meson setup _build
ninja -C _build
meson compile build-wordlist-index
meson compile -C _build build-wordlist-defs
meson compile build-wordlist-gresource-xml
meson compile build-wordlist-gresource

This builds the definitions and wordlist into one gresource file that can be loaded in to crosswords.

alt text

Load the resulting list in GNOME Crosswords and check a few entries.

crosswords/_build

alt text

*Closing Thoughts

With a config-driven workflow and cleaner tooling, adding new languages to GNOME Crosswords is now much easier. Swahili is just the beginning — this approach can scale to other underrepresented languages too.