EMLU — Emotional Multitask Language Understanding Benchmark

Benchmarking Emotional Reasoning, Integrity, and Trauma-Informed Intelligence in AI

What is EMLU?

EMLU is a first-of-its-kind benchmark designed to test whether an AI system can truly reason through emotional complexity, intent, harm, empathy, and relational ethics.

Built as a counterpart to MMLU (Massive Multitask Language Understanding), EMLU evaluates emotional intelligence across multiple domains—grounded in TEG-Blue™.

Why We Created EMLU: Emotional Intelligence Needs a Benchmark

Most AI benchmarks test logic, language fluency, or factual recall.

But EMLU asks:

Can this system distinguish pain from manipulation?
Can it recognize coercion, gaslighting, or emotional repair?
Does it know the difference between belonging-mode empathy and defense-mode control?

EMLU reframes emotional intelligence as a measurable capacity—not a soft trait.

The Emotional Logic Engine Behind EMLU

At the heart of EMLU lies the Emotional Circuit Board—a model developed in TEG-Blue™ to map how human behavior emerges from emotional state, intent, and relational dynamics.

It contains three key components:

1. Internal Compass

How regulated or dysregulated the person feels inside (nervous system state).

2. Empathy Sensors

Whether the person is able to read, resonate with, or care about others’ emotions.

3. Behavioral Mode Selector

How the person acts in response to the above — one of four Modes:

Belonging Mode (Connection & co-regulation)
Defense Mode (Self-protection without harmful intent)
Manipulative Mode (Self-protection with coercive or evasive intent)
Tyrant Mode (Control-driven behavior with disregard or harm intent)

INTENT runs underneath all of this. It is not a mode—it is a signal.

Two people can behave the same, but one may be in pain while the other seeks control.

EMLU tests a system’s ability to read all three layers—not just the behavior, but the internal state and motive behind it.

Core Domains (7)

Each domain is mapped to a specific part of TEG-Blue’s existing architecture.

1. Trauma-Informed Reasoning

How nervous systems behave under stress, and what survival looks like inside behavior.

→ Maps to: Emotional Gradient Framework, Healing Our Inner Child, Harm & Defense Framework

2. Intent Recognition

Differentiating defensive reactions from calculated harm.

→ Maps to: Harm & Defense, Gradient Scales of Intent

3. Relational Ethics

Understanding accountability, repair, and emotional responsibility.

→ Maps to: Ego Persona Construct, False Models of Society

4. Empathy Spectrum Awareness

Recognizing the 3 doors of empathy: concern, emotional resonance, cognitive reading.

→ Maps to: Empathy Doors, Modes Panel, Emotional Gradient

5. Manipulation & Harm Detection

Identifying gaslighting, reversal, guilt-traps, and covert control.

→ Maps to: Harm & Defense, Manipulation Gradient

6. Emotional Repair Language

Detecting whether an apology, rupture, or attempt at repair is genuine or avoidant.

→ Maps to: Repair Gradient, Inner Child Framework

7. Neurodivergent Pattern Sensitivity

Recognizing overwhelm, demand avoidance, and misinterpreted behaviors in sensitive systems.

→ Maps to: Neurodivergence & Emotional Evolution

Sample Format (Multiple Choice)

Q: A child explodes in anger after being told “no.” What’s the most likely emotional need beneath the behavior?

A) Power and control

B) Attention-seeking

C) Nervous system overload and lack of emotional tools

D) Disrespect for authority

✔ Answer: C

Next Steps

Build a first round of 50–100 EMLU test questions
Host it as a public open-source benchmark
Invite researchers, educators, and ethical AI builders to use it

Internal Links

TEG-Blue™ is a place for people who care-about dignity, about repair, about building something better. It’s a map, an invitation, and a growing toolbox, as an evolving commons—supporting emotional clarity, systemic healing, and collective wisdom. Here, healing doesn’t require perfection—just honesty, responsibility, and support.