Language IO Enhanced Chat for Salesforce | Toxicity Shield Overview & Configuration

Toxicity Shield protects agents and end users from translated messages that may be harmful or unpleasant, such as abusive, aggressive, or demeaning messages. It evaluates the toxicity of a translated message and can take action based on toxicity severity and your custom toxicity tolerances.

Enabling Toxicity Shield is a two-step process:

1. Configuring Toxicity Shield in the Customer Self-Service Portal
2. Enabling Toxicity Shield in Language IO Enhanced Chat for Salesforce

Configuring Toxicity Shield in the Customer Self-Service Portal

1. Log in to the Customer Self-Service Portal (CSSP).
2. In the top-nav, go to Integrations → "Toxicity Shield".
3. On this page, you can configure inbound and outbound message settings.
4. Under each section, you can:
   1. Set general and category-specific toxicity tolerance levels.
   2. Enable toxic translated message rewrite actions.
   3. Test toxic messages against your settings.
5. Click to expand each section and make changes.
6. Make sure to save changes by clicking "Save" in the lower-right corner of the page.
   

Understanding Tolerance Levels

General and category-specific tolerance levels can be adjusted between 0 (low toxic message tolerance) and 100 (high toxic message tolerance). The lower the tolerance level, the more sensitive Toxicity Shield will be to the translated message's toxicity. Toxicity Shield considers a translated message toxic if the message scores above your tolerance level for any category, general or specific.

Toxicity Severity & Indicators

In Language IO Enhanced Chat for Salesforce, Toxicity Shield assesses toxicity severity relative to your tolerance levels, tags translated messages with a colored circle representing the message's toxicity severity, and can take actions depending on severity and your settings. These indicators are only visible to agents, not end users.

🔴 high severity toxicity relative to your tolerance level.
🟠 medium severity toxicity relative to your tolerance level.
🟡 low severity toxicity relative to your tolerance level.

For inbound messages (end user → agent) in Enhanced Chat

If a translated message's toxicity severity is medium or high, the message is hidden beneath the eye (source) icon. If "Rewrite inbound toxic messages..." is enabled, the rewritten message is shown instead. The rewritten message is tagged with a sparkle (LLM) icon.

For outbound messages (agent → end user) in Enhanced Chat

If a translated message's toxicity severity is medium or high, the message is not sent, a "message-not-delivered" placeholder is added to the conversation with the source and translated message hidden beneath the eye (source) icon, and the agent is prompted to rewrite their message. If "Provide suggested rewrite (when possible)..." is enabled, a suggested revision is provided. If the agent accepts or edits and sends the revised message, and the translated message passes Toxicity Shield, then the translated message is sent. It is tagged with a sparkle (LLM) icon, and the previous toxic source and translated messages are nested under the eye (source) icon.

Configuring Toxicity Shield Rewrite Actions

Inbound (End User → Agent) Message Actions

Enable "Rewrite inbound toxic messages..." if Toxicity Shield should automatically revise toxic inbound message translations to decrease toxicity. For example:

Outbound (Agent → End User) Message Actions

Enable "Provide suggested rewrite (when possible)..." if Toxicity Shield should automatically suggest a revision of the source message to decrease translated message toxicity. In Language IO Enhanced Chat. Agents can choose to accept, edit, or discard the revision. For example:

Testing Toxicity Shield Tolerance Levels

Use the "Preview tool" under the respective inbound and outbound sections to test and tune your tolerance levels. The preview tool breaks down your message's toxicity scores across categories so you can compare them to your tolerance levels, indicates the message's toxicity severity, and provides a rewritten message if the corresponding setting is enabled.

1. Select the language of your message from the dropdown on the left.
2. Type your message.
3. Click the send icon on the right.
4. The tool shows the results below.
   

Enabling Toxicity Shield in Language IO Enhanced Chat for Salesforce

After configuring Toxicity Shield in the Customer Self-Service Portal, you must enable the feature in your Salesforce org.

1. Ensure that you have installed the latest version of Language IO Enhanced Chat for Salesforce.
2. Go to Setup → Custom Settings → "Language IO Enhanced Chat Settings".
3. Enable the "Enable Toxicity Filter" setting.
4. Save.

Toxicity Shield is now enabled in Language IO Enhanced Chat. Translated messages will be scored, tagged, and rewritten according to your configuration.