Skip to main content
The XO Platform NLP interpreter analyzes user utterances for emotional tone and returns a score you can use to guide conversation flow. For example, if a user is angry, you can automatically route the conversation to a live agent. Access tone scores from the Context object or configure events from Conversation Intelligence > Events > Sentiment Events. Learn more

​
Tone Types

The Platform evaluates six possible emotions:
ToneDescription
angryAnger detected in the utterance.
disgustDisgust detected.
fearFear detected.
sadSadness detected.
joyJoy detected.
positiveA special tone that evaluates the general positivity of an utterance.
Tones are not mutually exclusive — an utterance can score high for fear and mild for sadness simultaneously.
Starting from v8.1, the Platform detects emojis in user utterances and sets tone accordingly.

​
Tone Score Scale

Tone emotions are scored on a scale of -3 to +3:
ScoreMeaning
+3User definitely expressed the tone.
+2User expressed the tone.
+1User likely expressed the tone.
0Neutral — tone not expressed or suppressed.
-1User likely suppressed the tone.
-2User suppressed the tone.
-3User definitely suppressed the tone.
Positive values indicate the tone is explicitly expressed; negative values indicate it is explicitly negated. Examples:
  • “I am happy about this news” → positive joy score
  • “I am not happy about this news” → negative joy score

​
How Scores Are Calculated

The score is calculated from the base tone value and any modifiers (adverbs or adjectives that amplify or reduce the base tone). Examples:
  • “I am extremely disappointed” → higher angry score than “I am disappointed”
  • “I am not disappointed” → negative angry score
The tone analyzer compiles all base tones per emotion and calculates:
  • Current node score → stored in message_tone
  • Session average score → stored in dialog_tone (reset at end of each session)

​
Context Object Variables

VariableScopeDescription
message_toneCurrent nodeArray of tone emotions and scores for the current dialog node.
dialog_toneSessionArray of average tone emotions and scores for the entire conversation session.
Both variables return arrays of objects with tone_name, count, and level fields. Key/value pairs are only returned when a tone is detected. A level of 0 is returned when the tone is recognized as neutral. When accessing tone variables, handle positive, negative, zero, and undefined values.

​
Examples

message_tone
   0: tone_name: positive, level: 2
   1: tone_name: disgust, level: -2
   2: tone_name: angry, level: -2

dialog_tone
   0: tone_name: angry, level: -3
   1: tone_name: sad, level: -3
   2: tone_name: positive, level: 3
   3: tone_name: joy, level: 3
Sentence: “I don’t think that this is a good idea and I am not happy with how it came out, especially because of your attitude.” 
"dialog_tone": [
  { "tone_name": "joy", "count": 1, "level": 0.67 },
  { "tone_name": "sad", "count": 1, "level": 0.5 },
  { "tone_name": "angry", "count": 1, "level": 0.5 }
]
Sentence: “This is a great idea! I’m super excited already.” 
"dialog_tone": [
  { "tone_name": "joy", "count": 1, "level": 3 },
  { "tone_name": "sad", "count": 1, "level": 2.8 },
  { "tone_name": "angry", "count": 1, "level": -3 }
]
Sentence: “This was a funny and casually well-written book, a good read. But it’s a little frustrating because it abandons the narrative without finishing it.” 
"dialog_tone": [
  { "tone_name": "joy", "count": 1, "level": 1.5 },
  { "tone_name": "sad", "count": 1, "level": -1.5 },
  { "tone_name": "angry", "count": 1, "level": -1 }
]
Using tone scores in transition conditions: 
if context.message_tone.angry > 2.0
    then goTo liveAgent
For more information, see Context Object.

​
Adding Sentiment Words to Concepts

You can extend tone detection by adding custom words to tone concepts during concept training. Concept name syntax: ~tone-<tonename>-<level>
  • <tonename> — one of the 6 tone types listed above.
  • <level> — a number from 1 to 7, where 1 = -3, 4 = 0 (neutral), 7 = +3.
Note: Only add base tone words. Intensity modifiers like very or extremely are handled automatically by the Platform.
Examples:
WordConceptMeaning
freaking~tone-angry-7Very strong anger (+3)
Yikes!~tone-angry-5Mild anger (+1)
please~tone-angry-4Neutral tone (0)
Thanks!~tone-angry-1Not angry at all (-3)