GoldenDict: a gold mine of dictionaries

As a translator, I’ve used GoldenDict for at least a decade and a half as my central lookup dashboard: a single search across stacked dictionaries, with a convenient keyboard shortcut to look up selected words from anywhere. The original project spent years almost dormant. Only recently did I discover it has returned to stable releases: 1.5.0 dropped in May 2023, its first in over a decade, followed by 1.5.1 in May 2025.

Meanwhile, GoldenDict-ng came along, an active Qt6 fork with builds for Windows, macOS, and Linux (via Flathub, io.github.xiaoyifang.goldendict_ng). If you’re starting out today, it’s the one I recommend. The setup below works for both.

Formats and free sources

GoldenDict reads DSL (ABBYY Lingvo), StarDict, MDict, XDXF, and Zim, among other formats. It also reads .bgl, the format of the old Babylon premium dictionaries: if you held on to those files, you can use them again without having Babylon installed.

Free sources worth checking out:

• FreeDict: free bilingual dictionaries in StarDict format, including English–Portuguese and other pairs with Portuguese.
• Wiktionary: Wiktionary-Dictionaries offers ready-made StarDict versions. kaikki.org provides structured data in JSON, great as raw material (see the last section).
• DSL: lots of dictionaries circulate in this format. It’s worth checking the license before downloading.

Groups

Under Edit → Dictionaries (F3):

1. In the Sources → Files tab, point to the folders where your dictionaries live and check “Recursive”.
2. In the Groups tab, create groups by use, such as “EN→PT”. Drag the dictionaries in and sort them by priority. The Auto groups button creates language-pair groups for dictionaries that have that information built into their metadata.
3. Pick the active group from the selector in the toolbar.

Hunspell (morphology)

Hunspell lets GoldenDict find the base form of inflected words: “ran” leads to “run”, and the Portuguese “fizéramos” (we had done) leads to “fazer” (to do). It also suggests spellings when you make a typo.

1. Download the .aff + .dic pairs. The ones from LibreOffice do the job: pt_BR (VERO) and en_US/en_GB.
2. Under Sources → Morphology, point to the folder and check the languages.
3. Add morphology to each relevant group, preferably at the top.

Ivo is the man

For English-to-Portuguese translation, the standout is the English-Portuguese Translator’s Dictionary, by Ivo Korytowski. It has more than 46,000 entries gathered over more than 35 years of translating: idioms, technical terms, acronyms, and rare senses that ordinary dictionaries ignore. It’s free and comes in .bgl, ready for GoldenDict.

Converting formats with PyGlossary

PyGlossary converts between dozens of formats, through a graphical interface or from the command line:

pyglossary dicionario.bgl dicionario.ifo      # BGL → StarDict
pyglossary glossario.txt glossario.ifo        # Tabfile (term<TAB>definition) → StarDict

It reads DSL, MDict, and XDXF, but doesn’t write them. For output, the useful formats are StarDict, Tabfile, and BGL itself.

Building your own DSL

DSL is plain text with simple markup. That means you can generate compatible dictionaries from glossaries, spreadsheets, translation memories, or dumps like the ones from kaikki.org:

#NAME "Glossário jurídico EN-PT"
#INDEX_LANGUAGE "English"
#CONTENTS_LANGUAGE "Portuguese"

hearing
    [m1][p]n[/p] [trn]audiência[/trn][/m]
    [m2][ex]The hearing was adjourned.[/ex][/m]

The headword sits in column zero and the body is indented. The practical route is a Python script that reads the source (CSV, TMX, JSON) and writes the DSL; then dictzip produces a compact .dsl.dz.

The same process works for converting, for personal use, a digital dictionary you already own. Mobi and epub files are containers (HTML/XHTML plus resources), and the ebook-convert tool from Calibre turns them into an intermediate format that’s easier to work with. If the file is already structured as a dictionary, with a headword and a definition in each entry (like those that follow the Kindle or StarDict dictionary format), the main job is writing the parsing script for that format: identify the headword tag and the definition block and, with BeautifulSoup or lxml, loop through the entries and generate the DSL, converting the HTML tags into their equivalents ([b], [i], [ref] etc.) or simplifying everything to plain text. After that, it’s just a matter of compiling and testing.

In both cases, modern AI does the heavy lifting: writing the parser, mapping fields to tags, validating markup, and ironing out quirky edge cases.