Language Support

On This Page

AI opens up a world of possibilities when it comes to communication, notably across languages. Rather than needing to find an individual who can read, write, and speak the target language, you can now use AI to bridge that gap.

This topic discusses what to expect when using AI with multiple languages, and how to get the most predictable results using Appian's AI features across languages.

Tip: The best way to gauge the success of processing other languages is to gather examples and test them thoroughly, especially using examples that represent what you expect to encounter when the models are deployed to production. Learn more about what makes a good dataset.

Using AI to work with non-English languages

If your business operates in multiple locations or markets, you're likely familiar with both the opportunities and challenges that cross-language communication can present. Global business operations offer your goods or services to more markets, but you need to ensure that language barriers don't slow things down.

Appian's AI features can help efficiently process content in other languages, but it's important to remember that AI is non-deterministic, meaning that its output is never guaranteed. That's why one of our foundational AI principles is to treat AI as a helper, and to keep a human being in the loop to verify output when it makes sense. This is true when using AI features to analyze, extract, or create content in all languages, including English.

Note: If you're using plug-ins to convert documents from one format to another, refer to that plug-in's documentation in its AppMarket listing for more information.

Which languages does AI perform best with?

AI models are trained on multilingual data. However, their instruction-following and safety training is primarily written in English. As a general rule, AI provides the best results when you use languages with characters close to those in English and Latin.

As you work with your documents and other text-based content, you may observe the following languages have varying success:

Best	Good	Moderate	Challenging
English Italian Portuguese Dutch French Romanian Spanish	Czech German Hungarian Indonesian Polish Slovakian	Greek	Arabic Korean Chinese (Simplified) Japanese Hindi Bengali Swahili Yoruba

Think of language as a characteristic of your document, like handwriting or scan quality. Every characteristic influences how well AI is able to execute its task. To help you anticipate how well AI is able to process your documents, assemble a representative dataset and thoroughly test how it works with your non-English language content.

Feature-specific notes

Chat experiences

AI chat experiences can interpret questions and provide answers in many languages, in addition to English. As with all AI, you may experience a range of responses depending on factors like language, phrasing, and complexity of the request.

If you ask a question that references a field value, we recommend wrapping that literal value in quotation marks so AI is able to reference it. This might mean the field value you reference doesn't always match the language of your question. For example, if your record data is in English, you would format a question in Spanish as: ¿Cuantos casos están "Open"?. If the record field value was also in Spanish, you would format this question as: ¿Cuantos casos están "Abierto"?

To ensure the results are satisfactory when using AI to interact with non-English documents, such as in the documents chatbot component and Enterprise Copilot, we recommend testing both non-English documents and questions.

AI skills

Language support for AI skills depends on how the skill processes documents.

Supported OCR languages

If Appian or Amazon Web Services is the active OCR provider, they both use Amazon Textract and support the following languages:

Afrikaans
Albanian
Arabic
Basque
Belarusian
Bosnian
Breton
Bulgarian
Chinese (Simplified)
Chinese (Traditional)
Croatian
Czech
Dutch
English
French
German
Hebrew
Hindi
Hungarian
Icelandic
Italian
Japanese
Korean
Luxembourgish
Macedonian
Norwegian
Polish
Portuguese
Romanian
Russian
Serbian (Cyrillic and Latin)
Slovak
Slovenian
Sorbian
Spanish
Swedish
Turkish
Turkmen
Ukrainian

For AI skills that use the active OCR provider, setting Microsoft Azure or Google Cloud Platform as the active provider may support different languages. Check your provider's documentation for their supported language list.

In the advanced IDP tools AI skill, the Search Queries tool type only processes queries written in English. Search Queries is supported when Appian or Amazon Web Services is the active OCR provider, but isn't available with Microsoft Azure or Google Cloud Platform.

Broader language support with Visual elements & text mode

Document-based generative AI skills set to Visual elements & text mode support most languages. Use this mode to process documents in languages not supported by the active OCR provider.

The following skills support this mode:

To use this mode, in the AI skill object, under Identify Document Type, select Visual elements & text.

Implementation

See Multilingual Prompting to learn how to create reliable, high-quality, multilingual AI features by using English prompts with Appian’s generative AI capabilities.

Feedback

Was this page helpful?