Within the IA Accelerator you can customize the language settings of the search index, search skillsets, and Azure OpenAI prompts. To do this you must create a language ENV file in the scripts/environments/languages folder. This documentation will explain how to set up your own language file.
The first step in configuring a language for the IA Accelerator is to create a new language ENV file in the scripts/environments/languages folder. Simply add a new file in this folder and remember the name of the file you create. In this example, we have created a language ENV file for the Greek language (el.env).
Once you have created your language ENV file, we need to map to this file from scripts/environments/local.env. In this file find the setting named DEFAULT_LANGUAGE and set the value to match the file name you created above without the extension.
export DEFAULT_LANGUAGE="el"The next step is to populate the required values of your custom language ENV file. You can get a template of contents from scripts/environments/languages/language.env.example. The required values are:
| Parameter | Description |
|---|---|
| PROMPT_QUERYTERM_LANGUAGE | The language that Azure OpenAI will be prompted to generate the search terms in. This is used in the natural language prompt so example values would be "English" or "Greek". |
| SEARCH_INDEX_ANALYZER | The analyzer that the search index will use for all "searchable" fields except "translated_text". Supported analyzers can be found at https://learn.microsoft.com/en-us/azure/search/index-add-language-analyzers#language-analyzer-list |
| TARGET_TRANSLATION_LANGUAGE | The language that the cognitive service will use to translate text to if required. Supported languages and associated codes can be found at https://learn.microsoft.com/en-us/azure/ai-services/translator/language-support |
NOTE: It is important that all parameters have a value. In some cases your language of choice may not be available for one or more options, you must choose a supported value. In our example, for the Greek language, not all values were supported in Greek so we used English where Greek was not supported.