Customize word segmentation

Learn how to customize word segmentation dictionaries through the Algolia dashboard.

Word segmentation, or decompounding, splits a compound word into its parts. For example, Algolia splits the Dutch word “eettafel” (dining table) into “eet” and “tafel”. The engine can then match the parts of a word, not only the whole word.

The decompoundedAttributes setting specifies which attributes to decompound in your , and in which languages. On the , decompounding applies when the index’s query language supports it.

Algolia provides a built-in segmentation dictionary for German, Dutch, Finnish, Swedish, Danish, and Norwegian. You can customize each one in the Algolia dashboard.

Each Algolia uses one segmentation dictionary per language. Your changes affect every index that has this query language and sets decompoundedAttributes.

Open a segmentation dictionary

The Words Segmentation tab is available only for languages that have a segmentation dictionary. For any other language, the tab stays turned off.

1. Go to the Algolia dashboard and select your Algolia .
2. In the left sidebar, select Search.
3. Go to the Dictionaries page.
4. Select the language you want to customize.
5. Select the Words Segmentation tab.

Inspect segmentation

1. Open a segmentation dictionary.
2. Search for a word.
3. Under Algolia segmentation, see how the engine reads the word.

The search box takes one word and strips any spaces. If the engine splits the word, the parts appear as tags. Otherwise, the engine keeps the word as is.

Create custom segmentation entries

Algolia may not split a compound word that contains a brand name or a proper noun, such as the Dutch “gyprocplaten” (drywall). Searches then match the whole word only, not “gyproc” or “platen”.

To split a word yourself:

1. Open a segmentation dictionary.
2. Search for the word.
3. Under Algolia segmentation, select Edit.
4. Select Split, then enter the parts.
5. Select Save, then select Review and Save.

A split needs at least two parts, and each part must be a subset of the original word.

Turn off segmentation for words

Algolia may split a word that your catalog needs whole, such as the German brand “Volkswagen” into “volks” and “wagen”.

To keep a word whole:

1. Open a segmentation dictionary.
2. Search for the word.
3. Under Algolia segmentation, select Edit.
4. Select Keep as is.
5. Select Save, then select Review and Save.

Delete custom segmentation entries

You can only delete your own segmentation entries. Algolia’s segmentation always remains, and your custom entry overrides it.

To delete an entry:

1. Open a segmentation dictionary.
2. Search for the word.
3. Select Delete.
4. Select Review and Save.

Upload and download customizations

Open a segmentation dictionary, then select Actions:

• To upload your own list, select Upload your segmentations list.
• To export what you added, select Download your segmentations list as JSON or as CSV.

Upload format

CSV

word,decomposition,language,objectID
gyprocplaten,"gyproc,platen",nl,1
volkswagen,"volkswagen",de,2

Don’t put a space after a comma. The dashboard matches the header row character for character, so language doesn’t match language, and the upload skips the rows it can’t match. In CSV, put the parts in one quoted, comma-separated list. To keep a word whole, give it one part equal to the word.

Manage segmentation programmatically

The dashboard isn’t the only way to customize a dictionary. Segmentation entries live in the compounds dictionary.

• Algolia CLI: the algolia dictionary entries commands
• API clients: save, search, delete, and clear dictionary entries
• Search API: add or delete dictionary entries, search dictionary entries