Research Protocol: AI Assisted Terminography

Spring Semester 2026

1. Project Management

Project Name

Domain

2. Corpus Building

Select File (TXT, PDF, DOCX)

Language

⚙ Project Settings

▼

Prompt Templates

Variables: {domain}, {term}, {eval_guidelines}, {guidelines}, {chunk}, {language}, {context_block}

Term Extraction

Zero-shot

Few-shot

Chain-of-Thought Enhanced

Symbolic Parser Rules

Enable/disable checks, adjust weights

LLM Parameters (Generation Parameters)

Adjust temperature, max tokens, top-p for experiments

Temperature 0 = deterministic, 2 = very random

Max Tokens Maximum response length

Top-p Nucleus sampling (1.0 = off)

3. Extraction & Experimentation

Select Model

Target File for Extraction

Other files are used as context

Actions

4. Results & Verification

Term	Zero-shot			Persona (ISO)			CoT			Actions
Term	No RAG	Keyword	Vector	No RAG	Keyword	Vector	No RAG	Keyword	Vector	Actions

5. Neuro-Symbolic Feedback Loop

The LLM generates a definition → the symbolic parser checks it → if it fails, sends feedback → the LLM rewrites. Repeat until success or max iterations.

Select Term

Prompt Strategy

RAG Mode

Max Iterations

6. Concept Relation Extraction

Extract IS-A, PART-OF, CAUSES relations between terms with symbolic validation (cycles, contradictions).

🔑 API Keys

🟢 OpenAI (required for RAG embeddings)

🔵 Google Gemini

🟠 Anthropic Claude

🟣 OpenRouter