{"paper":{"title":"MRKL Systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"MRKL systems combine large language models with external knowledge and discrete reasoning modules to address inherent LM limits.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Amnon Shashua, Barak Lenz, Dor Muhlgay, Ehud Karpas, Erez Schwartz, Gal Shachaf, Hofit Bata, Kevin Leyton-Brown, Moshe Tenenholtz, Nir Ratner, Noam Rozen, Omri Abend, Opher Lieber, Shai Shalev-Shwartz, Yoav Levine, Yoav Shoham, Yonatan Belinkov","submitted_at":"2022-05-01T11:01:28Z","abstract_excerpt":"Huge language models (LMs) have ushered in a new era for AI, serving as a gateway to natural-language-based knowledge tasks. Although an essential element of modern AI, LMs are also inherently limited in a number of ways. We discuss these limitations and how they can be avoided by adopting a systems approach. Conceptualizing the challenge as one that involves knowledge and reasoning in addition to linguistic processing, we define a flexible architecture with multiple neural models, complemented by discrete knowledge and reasoning modules. We describe this neuro-symbolic architecture, dubbed th"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We define a flexible architecture with multiple neural models, complemented by discrete knowledge and reasoning modules.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the interfaces between neural language components and discrete knowledge/reasoning modules can be made reliable enough to deliver net gains over monolithic LMs without introducing new failure modes at the boundaries.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MRKL is a modular neuro-symbolic architecture that integrates LLMs with external knowledge and discrete reasoning to overcome limitations of pure neural language models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MRKL systems combine large language models with external knowledge and discrete reasoning modules to address inherent LM limits.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4db53586f57fd1747b022651e035d8e599f1ce57e992e810daf8154e113c7055"},"source":{"id":"2205.00445","kind":"arxiv","version":1},"verdict":{"id":"b8533b3f-b13e-4a88-8077-f85ec93238a2","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T07:27:22.153857Z","strongest_claim":"We define a flexible architecture with multiple neural models, complemented by discrete knowledge and reasoning modules.","one_line_summary":"MRKL is a modular neuro-symbolic architecture that integrates LLMs with external knowledge and discrete reasoning to overcome limitations of pure neural language models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the interfaces between neural language components and discrete knowledge/reasoning modules can be made reliable enough to deliver net gains over monolithic LMs without introducing new failure modes at the boundaries.","pith_extraction_headline":"MRKL systems combine large language models with external knowledge and discrete reasoning modules to address inherent LM limits."},"references":{"count":23,"sample":[{"doi":"","year":2019,"title":"Devlin, J., Chang, M.-W., Lee, K. & Toutanova, K.BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingin Proceedings of the 2019 Conference of the North American Chapter of ","work_id":"00b22d1f-29ac-44d3-bf74-53f063f7fcc9","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2005,"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","ref_index":2,"cited_arxiv_id":"2005.14165","is_internal_anchor":true},{"doi":"","year":2021,"title":"& Shoham, Y.Jurassic-1: Technical Details and Evaluation 2021","work_id":"070df3c5-50c5-4ecc-96da-8eda8669e750","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"PaLM: Scaling Language Modeling with Pathways","work_id":"a94f3ef7-2c49-4445-93fe-6ec16aafd966","ref_index":4,"cited_arxiv_id":"2204.02311","is_internal_anchor":true},{"doi":"","year":2020,"title":"et al.Exploring the Limits of Transfer Learning with a Unified Text- to-Text Transformer.Journal of Machine Learning Research21, 1–67","work_id":"d84daa05-7815-47fc-a7b2-07f57b7f5ccd","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":23,"snapshot_sha256":"d55a7c18d81a4d8d0bb51283b730d8426d9af077ec367868f9b06cab7a260e63","internal_anchors":8},"formal_canon":{"evidence_count":3,"snapshot_sha256":"21da73a1f8baeafce09a0bfa9902f932c0ec68bce1e1bbd26a170692e6bed4e5"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}