# Question answering

Question answering (QA) is a computer science discipline within information retrieval and natural language processing that builds systems to automatically answer human questions in natural language, using structured knowledge bases or unstructured document collections.

Question answering (QA) is a computer science discipline within the fields of information retrieval and natural language processing (NLP) that focuses on building systems capable of automatically answering questions posed by humans in a natural language. A QA implementation, typically a computer program, may construct its answers by querying a structured knowledge base or, more commonly, by extracting answers from an unstructured collection of natural language documents. Examples of such document collections include reference texts, compiled newswire reports, Wikipedia pages, and other World Wide Web content.

## History

Early question answering systems include BASEBALL and LUNAR. BASEBALL answered questions about Major League Baseball over a one-year period, while LUNAR addressed questions about the geological analysis of rocks returned by the Apollo Moon missions. Both systems were highly effective within their chosen domains; LUNAR, demonstrated at a lunar science convention in 1971, answered 90% of questions posed by untrained users in its domain. These systems relied on hand-written knowledge bases and language techniques similar to those used in ELIZA and DOCTOR, the first chatterbot programs.

In the late 1960s and early 1970s, Terry Winograd developed SHRDLU, a question-answering program that simulated a robot in a toy "blocks world," allowing users to ask about the state of that world. Its strength lay in a very specific domain and simple physics rules that were easy to encode. During the 1970s, expert systems with narrower knowledge bases emerged, producing more repeatable and valid responses within their areas. These systems differed from modern QA in their reliance on expert-constructed knowledge bases rather than statistical processing of large text corpora.

The 1970s and 1980s saw advances in computational linguistics, leading to projects like the Unix Consultant (UC) at U.C. Berkeley, which answered questions about the Unix operating system with a comprehensive hand-crafted knowledge base, and LILOG, a text-understanding system for tourism information in a German city. Although these systems remained at demonstration stage, they contributed to theories of computational linguistics and reasoning. Specialized systems, such as EAGLi for health and life scientists, were also developed.

In 2011, IBM's Watson, a QA computer system, won exhibition matches of Jeopardy! against Brad Rutter and Ken Jennings. Facebook Research released DrQA, an open-source system using Wikipedia as a knowledge source. The open-source framework Haystack by deepset combines open-domain QA with generative question answering and supports domain adaptation of language models for industry use. Large language models (LLMs) like GPT-4 and Gemini are successful QA systems enabling more sophisticated text understanding and generation, and when coupled with multimodal systems that process text, images, and audio, they significantly improve QA capabilities.

## Types

QA research addresses a wide range of question types, including fact, list, definition, how, why, hypothetical, semantically constrained, and cross-lingual questions. Reading comprehension QA, where a system answers questions about a given article, is a simpler form because the context is relatively short. Closed-book QA refers to systems that have memorized facts during training and can answer without explicit context, similar to humans taking closed-book exams.

Closed-domain QA deals with questions within a specific domain, such as medicine or automotive maintenance, and can exploit domain-specific knowledge often formalized in ontologies. It may also refer to situations where only limited question types are accepted, such as descriptive rather than procedural questions. Medical QA systems have been constructed, for example, for Alzheimer's disease. Open-domain QA handles questions about nearly anything, relying on general ontologies and world knowledge, and typically has much more data available for answer extraction.

## Modern Approaches

Modern QA systems increasingly leverage [large language models](https://www.wikiprompt.org/wiki/large-language-model) and [transformer](https://www.wikiprompt.org/wiki/transformer) architectures. These models, trained on vast text corpora, can perform closed-book QA by generating answers from learned knowledge. Open-domain systems often use retrieval-augmented generation, combining a retriever that selects relevant documents from a corpus with a generator that formulates the answer. This approach is used in frameworks like Haystack and DrQA.

[Deep learning](https://www.wikiprompt.org/wiki/deep-learning) and [neural networks](https://www.wikiprompt.org/wiki/neural-network) have driven recent advances, enabling systems to handle complex question types and multimodal inputs. [Generative AI](https://www.wikiprompt.org/wiki/generative-ai) models, such as those developed by [OpenAI](https://www.wikiprompt.org/wiki/openai) and [Anthropic](https://www.wikiprompt.org/wiki/anthropic), have demonstrated strong QA performance across diverse domains. Research continues on topics like interactivity, answer reuse, semantic parsing, answer presentation, knowledge representation, and social media analysis.

## Applications and Challenges

QA systems are applied in various fields, including customer support, healthcare, education, and information retrieval. They power virtual assistants, search engines, and enterprise knowledge management tools. Challenges include handling ambiguous questions, ensuring factual accuracy, and managing biases in training data. Current research explores embodied QA, where agents interact with physical environments, and visual QA, which combines image understanding with question answering.

## Future Directions

Future QA systems are likely to integrate more deeply with multimodal data, including audio and video, and to improve interactive capabilities, allowing systems to ask clarifying questions. Advances in reinforcement learning and [model efficiency](https://www.wikiprompt.org/wiki/model-pruning) may enable more robust and scalable systems. As [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) continues to evolve, QA will remain a central component of human-computer interaction.

## External links

- [Wikipedia: Question answering](https://en.wikipedia.org/wiki/Question_answering)

---
Source: https://www.wikiprompt.org/wiki/question-answering
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-14T21:02:54.016599+00:00