# AI That Links Information Across Documents for Better Answers > This patent describes an AI system that improves answers to questions by linking and understanding information across several related documents, rather than just one at a time. - **Patent:** US 12657244 - **Original title:** Inter-document attention mechanism - **Owner:** Microsoft Technology Licensing - **Granted:** 2026 - **Status:** Active - **Times cited:** 0 - **Field:** software, ai_ml, telecommunications, consumer_electronics ## What it does This technology improves how AI systems answer questions by understanding connections between different documents. When you ask a question, the system first finds several related documents. It then feeds these documents into a special AI model called a transformer, which is designed to 'pay attention' to important words and phrases. Crucially, this model propagates attention from words in a first document to words in a second document (claim 1), building a richer understanding. For example, if you ask about a person, and one document mentions their birth city while another mentions their current job, the AI can connect these details by letting the information flow between the documents. This process creates 'contextualized embeddings' for each word, allowing the AI to select the most relevant words to form an accurate answer to your query (claim 1). ## What it does NOT cover - Does not cover systems that do not use a transformer-based machine learning model for propagating attention between documents. - Does not cover methods that only propagate attention within a single document, without explicitly linking information between separate documents. - Does not cover systems where the identification of related documents is not based on similarity to the query or explicit links between documents (claim 2). - Does not cover AI models that retrieve answers without first processing tokens from multiple result documents through stacked network layers (claim 1). - Does not cover systems that fail to propagate attention between the query's own words and the words within the retrieved documents (claim 5, 6). ## The clever bit The clever part is explicitly propagating 'attention' from words in one document to words in another document, rather than treating each document as a separate island. This allows the AI to build a more complete understanding of a topic by connecting related pieces of information spread across different sources. ## Real-world examples 1. Advanced search engines like Bing or Google Search 2. Question-answering systems in customer service bots 3. AI assistants that summarize research papers 4. Enterprise search tools for large document repositories ## Why it matters This patent addresses a core challenge in natural language processing: understanding context across multiple sources. By enabling AI models to propagate attention between documents, it allows for more comprehensive and accurate answers to complex questions. This is crucial for search engines, virtual assistants, and knowledge management systems that need to synthesize information from vast amounts of text. It helps these systems move beyond simple keyword matching to provide more intelligent, nuanced responses. ## Frequently asked questions ### What does AI That Links Information Across Documents for Better Answers cover? This patent describes an AI system that improves answers to questions by linking and understanding information across several related documents, rather than just one at a time. ### Who owns patent US 12657244? Microsoft Technology Licensing owns this patent, granted in 2026. ### When does this patent expire? This patent is expected to expire on May 10, 2044, when the invention enters the public domain. ### What problem does this patent solve? This patent addresses a core challenge in natural language processing: understanding context across multiple sources. By enabling AI models to propagate attention between documents, it allows for more comprehensive and accurate answers to complex questions. This is crucial for search engines, virtual assistants, and knowledge management systems that need to synthesize information from vast amounts of text. It helps these systems move beyond simple keyword matching to provide more intelligent, nuanced responses. ### What does this patent NOT cover? Does not cover systems that do not use a transformer-based machine learning model for propagating attention between documents. **Full plain-English explainer:** https://patentbrief.org/patent/us/12657244/inter-document-attention-mechanism-12657244 **Original patent:** https://patents.google.com/patent/US12657244 --- _Source: PatentBrief — https://patentbrief.org. Patent facts are from public records; the plain-English explanation is PatentBrief's._ ## Related patents Semantically similar inventions in the PatentBrief corpus: - [How AI Understands Answers Across Multiple Documents](https://patentbrief.org/patent/us/20240338414/inter-document-attention-mechanism) — Microsoft's 2024 patent on using AI to find answers by understanding how information in one document relates to information in another. - [How Computers Use Memory Networks to Answer Questions](https://patentbrief.org/patent/us/10664744/watson-question-answering-system-deepqa) — A method for AI to search through large amounts of stored information by repeatedly 'hopping' through memory to find the most relevant facts for answering a question. - [How AI Models Understand Language Using 'Attention'](https://patentbrief.org/patent/us/10452978/transformer-attention-mechanism) — This patent describes a neural network architecture, known as a Transformer, that uses a "self-attention" mechanism to process sequences of information, like words in a sentence, by weighing the importance of different parts of the input. - [Improving AI Language Understanding with Aligned Knowledge Graphs](https://patentbrief.org/patent/us/11798529/generation-of-optimized-knowledge-based-language-model-through-knowledge-graph-m) — This patent describes a method for making AI language models smarter by combining them with knowledge modules, which learn from two different, but related, knowledge graphs. - [How AI Learns New Tasks Using Old Data Labels](https://patentbrief.org/patent/us/11062228/gpt-3-few-shot-learning) — A method for helping AI models understand new topics by grouping similar labels from different datasets into a shared, broader category.