Skip to main content
Back to External NLU Integrations Use these tools to validate that your External NLU adapter is working correctly and to understand intent detection behavior.

​
Test in Talk to Bot

After completing Adapter Configuration and enabling the Dialogflow API:
  1. Open your app and click Test.
  2. Enter a user utterance that matches an intent configured in Dialogflow Essentials.
  3. In the Debug Log panel, confirm the message: intent identified using External NLU is initiated.
Debug Log - intent identified using External NLU is initiated

​
Debug Log

The Debug Log shows whether intent was detected by External NLU before the Intent node initiated entry. If the API call fails, an error response appears in the log.
  • Intent detected by the external NLU engine is always marked as a Definitive Match. Intent detected by the External NLU engine
  • If intent is not detected by External NLU, this is also tracked in the Debug Log.
  • Hover over the information icon to view the K-Traceid and X-Traceid for each interaction. K-Traceid tracks and debugs platform activities; X-Traceid is a unique per-message identifier included in platform logs. K-Traceid and X-Traceid
  • Under Session Context & Variables, inspect intent and entity variables in detail. Session Context & Variables Custom Variables

​
Utterance Testing

Utterance Testing provides NLP Analysis for individual utterances. When intent is detected by an external engine, the utterance is not sent to the Kore.ai NLP engine, and the Matched Intents and Fields/Entities tables are not shown.
  • Intent detected by external engine: Intent detected by external engine
  • Click View Response to see the raw API response from the external NLU engine (available whether or not intent was detected). View the API response in JSON format
  • Intent not detected by external engine — Kore.ai NLU acts as fallback. NLP analysis shows results from all three Kore.ai engines plus ranking and resolver. See Ranking and Resolver. Intent not detected by external engine
  • Intent detected but not mappable — Kore.ai NLU acts as fallback. NLP analysis displays a message with the API response containing detected intents and parameters. Intent detected but cannot be mapped

​
Batch Testing

Test results (TP, FP, TN, FN) are based on the combined outcome of External NLU and Kore.ai NLU:
ResultCondition
TPExpected intent detected by either engine
FPWrong intent identified by the net result
TNNo intent detected by any engine (none expected)
FNNo intent expected, but some intent was detected
External NLU scores are not included in comparisons — there is no common scoring baseline across engines.
In the Matched Intent Score column of the downloaded CSV, intents detected by External NLU appear as ExtNLU (no score) instead of ML, FM, or KG scores. ExtNLU Matched intent score If External NLU returns multiple intents, thresholds are evaluated for all of them before concluding the test result. See Batch Testing for more details.

​
Health and Monitoring

Go to Test Cases > Detailed Analysis to see intents and entities identified by External NLU (displayed without scores). The Traits engine operates independently of external NLU.
When external NLU is enabled, some Health and Monitoring recommendations may not apply — recommendations are based on training data in the Kore.ai platform, not the external engine.
Intents and Entities identified by External NLU See Health and Monitoring for more details.

​
NLP Insights

NLP Insights captures intent identified and not-identified utterances with their NLP analysis. Failed External NLU API call details appear in the debug logs of NLP Insights. Failed External NLU API calls See NLP Insights for more details.

​
Conversation Behavior

BehaviorDetails
Intent triggeringWhen an intent is detected by the external engine, the corresponding Kore.ai task is triggered based on name mapping (case-insensitive).
Detection qualityDepends on training and capabilities of the external engine.
FallbackIf external NLU fails to detect an entity, Kore.ai NLP engine is used as fallback.
Entity orderingDefine entity nodes in the same sequence on both the platform and the external engine.
Interruption handlingOnly one behavior is supported: Switch to a new task without notification and discard the current task. See Interruption Management Settings.
API failureIf External NLU API returns an error or is unreachable, a fallback intent activates with the response: Error in continuing the conversation due to incorrect bot definition. See Standard Responses.
The platform handles all other conversation behaviors, including:
  • Re-prompts, error prompts, message responses, widgets
  • Channel support and channel-specific customizations
  • Authorization for service calls, alerts, small talk
  • Language detection, sentiment analysis, traits identification
  • Sub-intent and follow-up intent scanning decisions
  • Interruption handling, entity/intent priority settings
  • PII masking/redaction
  • Dialog task node execution, context object construction
  • Response messages, fulfillment, and logical flows
Response messages and fulfillment data from the external engine are ignored.
Intent and entity mapping uses case-insensitive name comparison between the two platforms.

​
Follow-up and Sub-Intent Identification

  • When all required parameters are available in context, follow-up intent scope activates on External NLU. Once active, utterances are sent to External NLU for intent identification.
  • When sub-intent scope is active on the Kore.ai platform, utterances are sent to External NLU for intent detection along with the parent intent context.