SnowPro Specialty: Gen AI Question 105
Single answerTRANSLATEA global support team stores customer emails in a Snowflake table and wants to normalize all incoming messages into English before downstream sentiment analysis and case routing. The source language varies by row and is not stored in a separate column. The team wants a solution that works directly in SQL with minimal pipeline changes. Which approach is the most appropriate?
- A
Use SNOWFLAKE.CORTEX.TRANSLATE(email_text, '', 'en') so Snowflake detects the source language for each row and translates the content into English.
- B
Use SNOWFLAKE.CORTEX.COMPLETE with a prompt asking for translation, because TRANSLATE requires the source language to be manually supplied and cannot infer it.
- C
Use the SQL TRANSLATE string function to convert the text to English, because it is optimized for multilingual text normalization in Snowflake.
- D
First call SNOWFLAKE.CORTEX.SENTIMENT on each email to infer the language, then pass that output into SNOWFLAKE.CORTEX.TRANSLATE.
Show answer and explanation
Correct answer: A
Explanation
For multilingual text stored in Snowflake, the most appropriate SQL-native solution is SNOWFLAKE.CORTEX.TRANSLATE. It is purpose-built for language translation and supports scenarios where the source language varies by row. When the source language is not known ahead of time, TRANSLATE can detect it, which minimizes upstream pipeline changes. This makes it a better fit than generic generation functions such as SNOWFLAKE.CORTEX.COMPLETE, which may be able to translate via prompting but are not the recommended first choice for a deterministic translation workflow. It is also important to distinguish Cortex TRANSLATE from the standard SQL TRANSLATE function, which only performs character substitution and is unrelated to multilingual AI translation. Snowflake documentation for Cortex AISQL functions describes TRANSLATE as the dedicated function for translation use cases in SQL.
- A. Correct.
Correct. SNOWFLAKE.CORTEX.TRANSLATE is designed for text translation in SQL. It supports automatic source-language detection when the source language argument is left empty, making it well suited when each row may contain a different language and the pipeline does not already track source language. Translating to English before downstream analysis is a practical pattern for standardizing multilingual inputs.
- B. Incorrect.
Incorrect. While SNOWFLAKE.CORTEX.COMPLETE can sometimes be prompted to translate text, it is not the most appropriate choice for this requirement. Snowflake provides a dedicated TRANSLATE function specifically for translation tasks, and TRANSLATE can handle source-language detection. Using COMPLETE for a specialized task introduces unnecessary prompt variability and is generally less direct than using the purpose-built function.
- C. Incorrect.
Incorrect. The SQL TRANSLATE string function is a standard character-mapping function, not an AI translation capability. It replaces characters based on positional mappings and cannot translate natural language text between languages. This is a common misconception caused by the shared function name.
- D. Incorrect.
Incorrect. SNOWFLAKE.CORTEX.SENTIMENT performs sentiment analysis, not language detection. It does not return a language code that can be reliably used as input for translation. Adding SENTIMENT here would not solve the stated problem and would introduce unnecessary processing.