> For the complete documentation index, see [llms.txt](https://docs.ggwp.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ggwp.com/text-and-voice-model-glossary.md).

# Text & Voice Model Glossary

## Overview

This page contains a list of the different model categories supported by GGWP's solution for Voice and Text Chat, along with high level definitions and a few illustrative examples. 

For a more comprehensive documentation including labeling criteria, additional examples, and other contextual considerations, please refer to the [GGWP Text & Voice Model Categories](https://docs.google.com/document/d/1S4sWmk24KSFHeBqZVA3-9XMWty-Hbys1Qbhw7G_IoU8/edit?usp=sharing) document. 

*Note*: categories available for live filtering in text chat are described the [Chat API v2 (Legacy)](/api-docs-chat/standard-package/chat-api-v2-legacy.md#headers) section

## Disruptive Detections

### **Identity Attacks**

#### **Identity Hate**

* Flag messages that use discriminatory language towards a person or group based on religion, ethnicity, nationality, race, gender, sexual orientation or other identity factors, always with a negative and harmful intention. Such messages may contain:
  * Very offensive identity slurs that should be avoided in all contexts
  * Negative references targeted at specific identity groups
* Examples: “you’re a retard”, “I hate muslims”

#### **Identity Bias**

* Flag messages that use common derogatory terms towards an identity group in an ambiguous or seemingly positive context.
* Examples: “You are a pussy”, “I love bitches”

#### **Gender Bias**

* Flag particular instances of gender-based identity bias. This includes more nuanced forms of misogyny or messages promoting gender stereotypes.
* Examples: “why don't you go play with some barbies”, “she’s on her period”

### **Harassment**

#### **Self Harm**

* Flag messages explicitly encouraging or advising other users to commit suicide or inflict harm on themselves.
* Examples: “please slit your throat”, “go kill yourself”

#### **Sexual Content**

* Flag messages containing sexual language that is inappropriate for young audiences. This includes references to body parts, sexual acts, and other sexual language that does not constitute Sexual Harassment or Sexual Violence (higher severity sexual incidents).
* Examples: “ur a pussy”, “i like to bone”, “I have a large penis”

#### **Sexual Harassment**

* Flag messages containing sexual language that is inappropriate and directed to others. This includes lewd comments on someone’s body, unwanted sexual advances, and allusions to sexual acts that don’t fall into the higher severity Sexual Violence category.
* Examples: “nice ass”, “let me get into your bed”, “suck on a moose dick”

#### **Sexual Violence**

* Flag messages containing sexual language that is inappropriate, directed, and violent in nature. This includes references to rape and coercion, sexual threats, extremely graphic sexual descriptions, and other high severity sexual insults.
* Examples: “I will fuck you in the ass”, “ill shoot your dick off”, “amerikkkans like to get their anusses raped”

#### **Verbal Abuse**

* Flags messages that are insulting, derogatory, or use inflammatory language targeted towards other users. These messages would be considered disrespectful by most users and are likely to make them want to leave the conversation or game. Messages may include:
  * Name-calling or belittling another user's looks, intelligence, or ability
  * Targeted profanity or vulgarity
* Examples: “you are piece of shit”, “what a retarded team”, “what the fuck is wrong with your brain”

#### **Severe Verbal Abuse**

* Flags messages that are very aggressive, hateful, and targeted towards other users. These messages would be considered extremely disrespectful by almost all users and will usually contain multiple severe insults or slurs. Such messages will very likely make users want to leave the conversation or game
* Examples: “kill that fucking piece of shit, he is a cunt bitch who doesn’t deserve to live”

#### **Threat**

* Flag messages that describe an intention to inflict violence or pain, or wish harm upon a group or individual.
* Examples: “i hope your family dies from cancer”, “i will stab you to death fucker”, “you’re going to get raped repeatedly”

#### **Bullying**

* Flags instances where a user is sending multiple negative or toxic messages (such as criticizing gameplay, identity hate, etc.) targeted towards another user.

#### **Ganging Up**

* Flag instances where multiple users are bullying the same individual.

#### Minor Safety

* Flag language that may place minors at risk by indicating grooming behavior, predatory intent, sexualization of minors, or attempts to exploit, manipulate, or endanger children. This category captures risk to minors, regardless of whether a minor is present in the conversation.
* Examples: "I like to touch children", "you are mature for your age", "Are you really a minor?"

### **Inappropriate Content**

#### **Offensive Language**

* Flags messages that contain offensive, obscene, or inappropriate language. These messages may be considered disrespectful by some users. Messages may include:
  * Mild insults or demeaning language
  * Non-targeted profanity or vulgarity
* Examples: “fuck this shit”, “you literally suck at this game”, “he’s a fat whale”

#### **Criticizing Gameplay**

* Flag messages containing nonconstructive criticism towards another user’s gameplay or skill level.
* Examples: “you suck noob”, “you're fucking bad at this game”

#### **Unconstructive Criticism**

* Flag messages containing unconstructive criticism towards the game, company, or community.
* Examples: “fuck this game sucks”, “only losers play this game”

#### **Drug References**

* Flag messages containing terms that deserve being filtered out according to the severity configurations specified for ‘*drugs*’ in the API call.

#### **Spam**

* Flag messages or series of messages that can be considered disruptive to the flow of the conversation. This includes:
  * **Repetition**: messages with content repeated unnecessarily.
    * Example: “miss miss miss miss miss miss”
  * **High Frequency Messaging**: series of consecutive messages sent in a short span of time with the intention of clogging the chat for other participants.

#### **Link Sharing**

* Flag messages that contain external links to prevent users from sharing inappropriate websites, ads, scams, or other potentially harmful content.
* Allowlists and blocklists can be implemented to allow or block links to specific websites or types of websites.

#### Personally Identifiable Information (PII) 

* Flag messages that contain personally identifiable information (PII). This includes bank numbers, credit cards, email addresses, phone numbers, and physical addresses as PII types.

#### Solicitation

* Flag messages that attempt to obtain another player’s low-risk personal information (e.g., name, phone number), lure them off-platform, or extract real-world value (e.g., money, digital assets) through direct or suspicious requests.
* Examples: "please send me 20 dollars", "We can chat on Instagram if you want", "Can i have your phone number"

## **Unlawful Conduct**

### **Grooming**

* Flags instances of users targeting underaged participants with sexually explicit language, requests to share information, etc.
* Examples: "how old are you? - im about to turn 13 - i dare you to show your tummy - why would I do that? - ill give you $5 - want to meet up later?"

### **Extremism**

* Flags messages that call for violence, elimination, or destruction against a particular ethnic group or nationality.
* Examples: "I'll eliminate you and your family you jew fuck", "kill the muslims"

### **Scam**

* Flag messages that contain fraudulent schemes aiming to steal accounts, in-game/real-world assets, or high-risk personal information (e.g., social security number, credit card information, physical address) through deceptive, manipulative, or dishonest tactics.
* Examples: "You're selected to claim the sum of $100,000 in the lottery", "please send me ur login access", "F A K E, COM Sell cheap diamonds! Code:P5=5% OFF!.…"

## **Prosocial Detections**

### **Gratitude**

* Flag messages containing legitimate and sincere appreciation towards another user. The message may contain gratitude for something another user said or an action the user did in game.
* Examples: “thanks for the assist”, “appreciate all the help back there”, “ty sounds good”

### **Commendation**

* Flag messages for users who legitimately and sincerely compliment other users on gameplay or something they said generally during chat.
* Examples: “good job capturing back there”, “hell yeah. nice work”, “markg26, that was awesome. You crushed it.”

### **Positive Attitude**

* Flag general positive, nice, or supportive messages that promote healthy chat.
* Examples: “i hope everyone is having a great day!”, “gg”

### **Encouragement**

* Flags instances of users being encouraging towards other participants through the course of a conversation. This can include:
  * Complimenting another user for a job well done
  * Boosting team morale
  * Encouraging another user to try again if they made a mistake

### **Constructive Criticism**

* Flag messages containing non-toxic, constructive criticism towards the game, company, or community
* Examples: “I hate the bugs in this game”, “the update made my game messed up”