AI 용어집
인공지능 완전 사전
Conversational Context Management
Ability of a QA system to retain and use the history of exchanges to understand implicit references and maintain response consistency across multiple dialogue turns.
Anaphora Resolution
NLP process that identifies and links pronouns or referential expressions (e.g., 'he', 'this solution') to their antecedents in the preceding conversational context.
Entity Tracking
Mechanism that keeps the values of entities (dates, names, places) up-to-date throughout the conversation, allowing their modification or reuse in subsequent questions.
Dialogue Policy
Decision-making component that, based on the current dialogue state, determines the next action to be executed by the system (ask a question, provide an answer, clarify).
Response Generator
Module that transforms the action decided by the dialogue policy into a natural and coherent sentence or message intended for the user.
Dialogue Turn
Atomic unit of a conversation, corresponding to a pair of exchanges (user utterance and system response) or to a single utterance in more complex models.
Question Rephrasing
Technique for rephrasing an ambiguous or incomplete question into a more precise query, often by relying on the context of previous turns.
Topic Change Detection
Ability of the system to identify when the user introduces a new topic, requiring a partial or total reset of the conversational context.
Conversational Memory
Storage architecture (short-term or long-term) for key information from a conversation, enabling interaction personalization and continuity across multiple sessions.
Interactive Clarification
A process where the system asks follow-up questions to the user to resolve ambiguity, get a missing detail, or refine its understanding before responding.
Conversational Language Model
A type of language model (e.g., GPT, BERT) specifically pre-trained or fine-tuned on conversational data to better handle the specificities of natural language in conversation.
Reasoning over History
The advanced ability of a QA system to not just retrieve past information, but to infer new knowledge by analyzing the logical sequence of exchanges.
Interruption Handling
A mechanism allowing the system to handle cases where a user interrupts a response, asks an off-topic question, or abruptly changes the subject, while trying to remain relevant.
Conversational Prompt Engineering
The art of designing instructions (prompts) for large language models by explicitly incorporating the conversation history to guide the generation of the next response.
Multi-turn System
Refers to a QA system capable of handling dialogues that span multiple exchanges, as opposed to single-turn systems that treat each question in isolation.