Supported languages

Support levels and key considerations

Language support for genAI components is divided into three levels. As language coverage expands, additional languages are expected to begin at levels 2 or 3.

Level 1

Languages have undergone thorough testing and evaluation, ensuring a robust and reliable experience. Experience Cloud genAI features are ready to be deployed with these languages, and Medallia considers them capable of delivering quality comparable to English.

Level 2

Languages show acceptable performance with partial testing, but not enough for full validation. Level 2 languages are most suited to early adopters and for development feedback. They may perform well in practice, but lack the rigor and validation of level 1 languages.

Level 3

Languages may be supported by the large language model, but they have little to no internal validation. Level 3 languages are enabled for preview only if the underlying model for the feature is expected to reasonably support it. These languages have not been validated, so high quality is neither expected nor guaranteed.

These are important factors when considering language support levels:

• Support levels indicate the depth of validation and accuracy as determined by third-party evaluations.

• Movement between levels of support is not automatic but requires both demand and data.

• Translation quality is not a separate support level; rather, it depends on the target language's level.

• Language dialects are not separately supported or leveled, though some components can be manually altered to use regional dialects, like custom instructions with Smart Responses or changing theme titles for Themes with GenAI.

Language data requirements

The following list describes language data input and output, translation support, and Text Analytics processing dependency for each genAI feature.

Smart Response

Initial data:

• Feedback received in native supported language
• Custom instructions may also be added to handle nuances in language support

Output: Smart Response email in native language matching original feedback language

Translation supported: No

Text Analytics data dependency: None

Intelligent Summaries for Text Analytics

Initial data:

• Reporting application language is set to a supported language
• Required Text Analytics processing in native or translated supported language
• Unsupported language input is not included in summary output:

Output: Summary in reporting application language

Translation supported: Yes

Text Analytics data dependency: Processed text in supported language required

Intelligent Summaries for conversational data

Initial data:

• Speech or chat transcript in native supported language

Output: Native language summary; summary output does not rely on reporting application language or processed language

Translation supported: No

Text Analytics data dependency: None

Root Cause Assist

Initial data:

• Reporting application language is set to a supported language
• Recommended Text Analytics processing in native or translated supported language

Output: Summary in reporting application language

Translation supported: Yes

Text Analytics data dependency: Processed text in supported language recommended

Themes with GenAI

Initial data:

• Required Text Analytics processing in native or translated supported language(s)
• Feedback in unsupported languages is not processed

Output: Theme in processed language

Translation supported: Yes

Text Analytics data dependency: Processed text in supported language required

Smart Topic Builder

Initial data:

• Required Text Analytics processing in native or translated supported language(s)
• Feedback in unsupported languages is not processed

Output: Topic name, definition components, rules, and word groups in processed language

Translation supported: Yes

Text Analytics data dependency: Processed text in supported language required

Insights Assistant

Initial data:

• User queries in English
• Verbatim comments and Text Analytics data in other languages are not filtered out

Output: Chat response in English

Translation supported: Yes

Text Analytics data dependency: Processed text in supported language recommended

Mixed-language input

Mixed-language input impacts only AI features that use verbatim inputs across records. This means that only the Intelligent Summaries for Text Analytics feature is impacted because it processes multiple comments that may be in different languages. Phrases processed in any supported language are included in the summary, but phrases in unsupported languages are ignored by the model.

Summaries are produced as long as the number of comments in any supported language meets or exceeds the summarization model's minimum sample size. Summaries do not include input from unsupported languages, and there are no quality impacts on the summary from data in unsupported languages because the model ignores that data. For more information about data eligibility for Intelligent Summaries for Text Analytics, see Configuration, role permissions, data eligibility, and languages.

For example, for a program with English, Spanish, and Italian comment data, summaries include only data from the English and Spanish comments, and summaries have the same quality as if the Italian comments were not present.

Language requests and development

To request improvements in language support or the addition of new languages, if you have access to help.medallia.com, submit a Product Enhancement Request and use the Product Area [Platform] AI Platform, Models, and Common Services. Include the language requirements, desired genAI feature(s), and associated contract to support prioritization and roadmap planning.

Strategic considerations for language & AI deployment

The effectiveness of AI-driven insights depends heavily on the predominant language of your data. Before implementation, use the information below to decide whether a multi-language program is necessary or if translation to English is more efficient.

1. The global Text Analytics language processing setting

   The global Text Analytics language processing setting is the "master switch" for your instance. Changing this has significant implications for existing data.

   • Theme generation — Themes are generated based on this global setting. If set to English, all themes are in English, even if the source data is in other languages.

   • AI training data — Most AI models are trained with English data, so using English data typically yields more accurate results.

   • Multi-lingual input — Intelligent Summaries for Text Analytics testing did not indicate a clear correlation between the percentage of English in the input on the accuracy of the output. A mixed language input may cause more variation.

2. Feature impact & dependencies

   • Smart Response — Governed by the native language of the received feedback. Output is generated in the matching native language, does not rely on or affect Text Analytics, and does not support translation.

   • Topics & Text Analytics summaries — Governed by the reporting application language. This can be adjusted based on the specific user's reporting preference.

   • Intelligent Summaries for conversational data — Governed by the record's native language. These are based on the transcript's original language; global settings do not override this.

   • Root Cause Assist — Governed by the reporting application language. Root Cause Assist will function as long as the reporting application language is set to a supported language.

   • Themes — Governed by the global Text Analytics language processing setting. This is centralized and affects both historical and new data.

   • Smart Topic Builder — Governed by the processed language. Generated topic names and rules are in English.

3. Key constraints & workarounds

   • Transcripts — Intelligent Summaries for conversational data are generated before Text Analytics processing. If a transcript is in Portuguese but the system expects English, the summary will not generate. This is a technical limitation.

   • Manual overrides — You can override the native language detection via Experience Program settings to force Text Analytics to process data in a specific language.