Customers can interact with AI agents in their preferred language across various channels. Webex AI Agent support multiple languages, allowing the AI agents to understand customer input and respond in the same language. You can initially create an AI agent in one language and enable additional languages as needed.

Webex AI Agent supports Polymatch and Language Encoder multilingual models for training AI agents. These models are available as part of the Swiftmatch NLU engine's settings. For an AI agent trained using Polymatch, a distinct model is created for each language and added to the AI agent. The AI engine performs inference based on the language of the customer query. The Language Encoder model performs best for training data on long utterances, while Polymatch does the same for short utterances. Note that the languages supported by Polymatch and the Language Encoder differ. For more information, refer to the following sections.

Supported languages for Scripted AI Agents (Answer questions)

You can use a single scripted AI agent for conversations in multiple languages across channels. Currently, only Swiftmatch NLU supports multi-languages to answer questions. For information about configuring multi-languages for AI agent, see the Add language(s) to AI Agent section in Scripted AI Agent for answering questions. The following table lists the languages supported for scripted AI agent to answer questions:

Language

Language code

Language Encoder & PolymatchMindmeldSpellcheck
AfrikaansafYes
AlbaniansqYes
AmharicamYes
ArabicarYesYes✓
ArmenianhyYes
AymaraayYes
AzerbaijaniazYes
BasqueeuYes
BelarusianbeYes
BengalibnYes
Berber languagesberYes
BosnianbsYes
BretonbrYes
BulgarianbgYesYes
BurmesemyYes
CatalancaYesYes
Central/Kadazan DusundtpYes
Central KhmerkmYes
ChavacanocbkYes
ChinesezhYes✓
Coastal KadazankzjYes
CornishkwYes
CroatianhrYesYes
CzechcsYes
DanishdaYesYes✓
DutchnlYesYes✓
Eastern MarimhrYes
EnglishenYesYes✓
EsperantoeoYes
EstonianetYesYes
FinnishfiYesYes✓
FrenchfrYesYes✓
GalicianglYes
GeorgiankaYesYes
GermandeYesYes✓
GreekelYesYes
HausahaYes
HebrewheYesYes
HindihiYesYes
HungarianhuYesYes
IcelandicisYes
IdoioYes
IndonesianidYesYes
InterlinguaiaYes
InterlingueieYes
IrishgaYesYes
ItalianitYesYes✓
JapanesejaYes✓
KabylekabYes
KazakhkkYes
KoreankoYesYes✓
KurdishkuYes
LatvianlvYes
LatinlaYes
Lingua Franca NovalfnYes
LithuanianltYesYes
Low German/SaxonndsYes
MacedonianmkYesYes
MalagasymgYes
MalaymsYes
MalayalammlYesYes
Maldivian (Divehi)dvYes
MarathimrYes
NepalineYes
Norwegian (Bokmål)nbYesYes✓
OccitanocYes
Persian (Farsi)psYesYes
PolishplYesYes✓
PortugueseptYesYes✓
RomanianroYesYes
RussianruYesYes✓
SerbiansrYes
SindhisdYes
SinhalasiYes
SlovakskYes
SlovenianslYes
SomalisoYes
SpanishesYesYes✓
SwahiliswYes
SwedishsvYesYes✓
TagalogtlYes
TajiktgYes
TamiltaYes
TatarttYes
TeluguteYes
ThaithYes
TurkishtrYesYes✓
UighurugYes
UkrainianukYesYes
UrduurYes
UzbekuzYes
VietnameseviYesYes
Wu ChinesewuuYes
Yue ChineseyueYes

Supported languages for Scripted AI Agents (Perform actions)

You can use a single scripted AI agent for conversations in multiple languages across channels. Currently, language encoder and Polymatch NLU supports multi-languages to perform actions. For information about configuring multi-languages for AI agent, see the Add language(s) to AI Agent section in Scripted AI Agent for performing tasks. The following table lists the languages supported for scripted AI agents to perform actions:

Language

Language code

Language EncoderMindmeldPolymatchSpellcheck
ArabicarYesYesYes✓
BulgarianbgYesYesYes✓
CatalancaYesYesYes✓
ChinesezhYesNo✓
CroatianhrNoYesYes✓
DanishdaYesYesYes
DutchnlYesYesYes
EnglishenYesYesYes✓
EstonianetYes
FinnishfiYesYesYes✓
FrenchfrYesYesYes✓
GeorgiankaYesYesYes
GermandeYesYesYes✓
GreekelYesYesYes
HebrewheYesYesYes✓
HindihiYesYesYes
HungarianhuYesYesYes
IndonesianidYesYesYes
IrishgaYesYesYes
ItalianitYesYesYes✓
JapanesejaYesNo✓
KoreankoYesYesYes✓
LithuanianltYesYesYes
MacedonianmkYesYesYes
MalayalammlYes
Mongolian mnNoYesYes
NepalineYes
Norwegian (Bokmål)nbYesYes✓
PolishplYesYesYes✓
PortugueseptYesYesYes✓
RomanianroYesYesYes
RussianruYesYesYes✓
SpanishesYesYesYes✓
SwedishsvYesYesYes✓
TurkishtrYesYesYes✓
UkrainianukYesYesYes
VietnameseviYesYesYes

The accuracy of a AI agent trained with Swiftmatch and mindmeld depends on various features that are involved in the training of the AI agent and the inference of incoming user messages. The key features that decide whether a language is supported and its accuracy are:

  • Text (support for the language and the script it is written in)
  • Contextual spellcheck (if spelling mistakes in the data or user query can be corrected)
  • Wordforms (handling various forms and synonyms for words present in training data)

For scenarios involving tasks and requiring the usage of system entities, support for system entities in the desired language is also required.

For entity recognition in AI agent to perform tasks:

  • Custom list, regex, and free-form entity types are supported in all the languages.
  • Support for Date, Time, Duration, Cardinals, Ordinals entity types indicated by the Common system entities column.
  • Other system entities supported for English only.

Table 1. Languages verses entities supported
LanguageOrdinalQuantityCardinalMoneyDurationDate/timePersonLocation
ArabicSupportedSupportedSupportedSupportedSupportedSupported
BulgarianSupportedSupportedSupportedSupportedSupportedSupported
CatalanSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
ChineseSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
CroatianSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
DanishSupportedSupportedSupportedSupportedSupportedSupported
DutchSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
EnglishSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
FinnishSupportedSupportedSupportedSupportedSupportedSupported
FrenchSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
GeorgianSupportedSupportedSupportedSupportedSupported
GermanSupportedSupportedSupportedSupportedSupportedSupportedSupported
GreekSupportedSupportedSupportedSupportedSupportedSupported
HebrewSupportedSupportedSupportedSupportedSupported
HindiSupportedSupportedSupportedSupportedSupported
HungarianSupportedSupportedSupportedSupported
IndonesianSupportedSupportedSupported
IrishSupportedSupportedSupportedSupportedSupportedSupported
ItalianSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
JapaneseSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
KoreanSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
LithuanianSupportedSupportedSupported
MacedonianSupportedSupportedSupportedSupportedSupportedSupported
MongolianSupportedSupportedSupportedSupportedSupportedSupported
Norwegian BokmålSupportedSupportedSupportedSupportedSupportedSupportedSupported
PolishSupportedSupportedSupportedSupportedSupportedSupported
PortugueseSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
RomanianSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
RussianSupportedSupportedSupportedSupportedSupportedSupported
SpanishSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
SwedishSupportedSupportedSupportedSupportedSupportedSupportedSupportedSupported
TurkishSupportedSupportedSupportedSupportedSupportedSupported
UkrainianSupportedSupportedSupportedSupportedSupportedSupported
VietnameseSupportedSupportedSupportedSupported