{"work":{"id":"27eaec54-c105-4969-8188-da5f0fca3688","openalex_id":"https://openalex.org/W4400526027","doi":"10.1145/3626772.3657717","arxiv_id":"2005.11401","raw_key":null,"title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","authors":null,"authors_text":"Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal","year":2020,"venue":"cs.CL","abstract":"Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, but have so far been only investigated for extractive downstream tasks. We explore a general-purpose fine-tuning recipe for retrieval-augmented generation (RAG) -- models which combine pre-trained parametric and non-parametric memory for language generation. We introduce RAG models where the parametric memory is a pre-trained seq2seq model and the non-parametric memory is a dense vector index of Wikipedia, accessed with a pre-trained neural retriever. We compare two RAG formulations, one which conditions on the same retrieved passages across the whole generated sequence, the other can use different passages per token. We fine-tune and evaluate our models on a wide range of knowledge-intensive NLP tasks and set the state-of-the-art on three open domain QA tasks, outperforming parametric seq2seq models and task-specific retrieve-and-extract architectures. For language generation tasks, we find that RAG models generate more specific, diverse and factual language than a state-of-the-art parametric-only seq2seq baseline.","external_url":"https://arxiv.org/abs/2005.11401","cited_by_count":14,"metadata_source":"pith","metadata_fetched_at":"2026-08-05T02:28:24.338817+00:00","pith_arxiv_id":"2005.11401","created_at":"2026-05-09T00:14:28.051609+00:00","updated_at":"2026-08-05T02:28:24.338817+00:00","title_quality_ok":true,"display_title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","render_title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks"},"hub":{"state":{"work_id":"27eaec54-c105-4969-8188-da5f0fca3688","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":199,"external_cited_by_count":14,"distinct_field_count":21,"first_pith_cited_at":"2020-05-28T17:29:03+00:00","last_pith_cited_at":"2026-07-09T01:08:33+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-08-22T11:59:24.438164+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":34},{"context_role":"method","n":3},{"context_role":"other","n":1}],"polarity_counts":[{"context_polarity":"background","n":32},{"context_polarity":"use_method","n":3},{"context_polarity":"unclear","n":2},{"context_polarity":"support","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","claims":[{"claim_text":"Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, b","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"SHIELD [157] Supervised Yes DARPA-E3, ATLASv2 Yes Prec DeepSeek-R1 LogReasoner [103] Supervised Yes BGL, Spirit No F1 Qwen2.5-1.5B Data Generation, Benchmarks & Visions Karlsen et al. [71] Supervised Yes CSIC, Thunderbird, BGLYes F1 RoBERTa/GPT LogEval [22] Supervised No BGL, Thunderbird No F1 GPT-4 AD-LLM [192] Semi-Sup Yes News, Reviews, Spam No AUC LLaMA-3/GPT-4o AnomalyGen [84] Supervised Yes Hadoop, HDFS Yes F1 GPT-4o 29 6.2.2 LLM-based Approaches for Anomaly Detection.Table 4-5 summarizes ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"RAGs contribute to addressing critical limitations such as hallucinations in LLMs' responses, knowledge cutoffs, especially when LLMs are required to deal with do- main-specific questions that require precise expertise , and , importantly for the sake of transparency in LLMs ' computations, generations, and outputs , the lack of traceability [1,2]. By dynamically retrieving relevant document excerpts at inference time, RAGs ena- ble LLMs to generate responses anchored in verifiable source materi","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"reproducibility, we provide the complete data of our experiments at https://github.com/lorenzbrehme/CARE. 1 Introduction Retrieval-Augmented Generation (RAG) is an ongoing trend in research and has been continuously developed over the past years [5]. It augments their knowl- edge through the integration of external data sources and addresses hallucina- tions in large language models (LLMs) [15]. A basic RAG system consists of three components [13]: (i) Indexing component-responsible for indexing","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"\"Hallucination\", the generation of plausible but factually incorrect content, remains a major barrier to enterprise LLM adoption [12], typically categorized into intrinsic (contradicting inputs) and extrin- sic (fabricating details) types [27]. Standard mitigation strategies operate at different stages of the model lifecycle: RLHF [22] aligns models during training, while Retrieval-Augmented Generation (RAG) [15] grounds inference in external evidence. However, gen- erators may still fail due to","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Sai utilizes a Multi-Agent System (MAS) architec- ture [7, 8] integrated with a daemon-based message management system. The MAS consists of specialized agents that collaborate to decompose and execute complex tasks, supported by base LLMs for foundational reasoning and natural language pro- cessing. To incorporate domain-specific expertise, Retrieval-Augmented Generation (RAG) [9] modules are utilized to feed external knowledge directly into the agents' decision-making and code-generation proces","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Unlike passive language models, agentic systems are dynamic, autonomous agents that reason, plan, and interact with external knowledge sources [16,17]. RAG combines pre-trained parametric memory with non-parametric memory (dense vector indices of documents) for knowledge-intensive tasks, enabling models to generate more specific, diverse, and factual language than parametric-only baselines [18]. They maintain task focus, retrieve information from external resources, synthesize diverse informatio","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (31 contexts).","role_counts":[{"n":31,"context_role":"background"},{"n":3,"context_role":"method"},{"n":1,"context_role":"other"}]},"error":null,"updated_at":"2026-05-19T14:41:31.247174+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"ab027a6a-af2b-4f6c-8e2f-4d3c57f079b0","orcid":null,"display_name":"Patrick Lewis"},{"id":"19b22360-f7f4-4cb9-b4fd-ab815d17f54a","orcid":null,"display_name":"Ethan Perez"},{"id":"5cf0b0f3-cf4a-4e65-a452-37fb8a08fbb8","orcid":null,"display_name":"Aleksandra Piktus"},{"id":"32209fff-0763-442b-a3eb-fc7c1b841804","orcid":null,"display_name":"Fabio Petroni"},{"id":"bffad31d-14ea-461f-adaf-4b6ed63eec65","orcid":null,"display_name":"Vladimir Karpukhin"},{"id":"63b793df-28cb-4279-a2b8-3385a04bd3d4","orcid":null,"display_name":"Naman Goyal"}]},"error":null,"updated_at":"2026-05-19T14:41:32.271208+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T08:58:01.028306+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":17},{"title":"Reflexion: Language Agents with Verbal Reinforcement Learning","work_id":"778f739e-5f55-4961-8a2a-e4736a2757f4","shared_citers":13},{"title":"Retrieval-Augmented Generation for Large Language Models: A Survey","work_id":"b80d2790-6cd9-4c87-b3c4-de404f99a80e","shared_citers":13},{"title":"AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation","work_id":"92b7eb9c-c3d8-4518-a376-06fa15dd895b","shared_citers":10},{"title":"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena","work_id":"d0c30cd7-81e1-4159-a87f-f6adca77ff08","shared_citers":10},{"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","shared_citers":9},{"title":"MemGPT: Towards LLMs as Operating Systems","work_id":"2698f5ad-c84c-40ca-b839-0912dae10ba2","shared_citers":9},{"title":"Toolformer: Language Models Can Teach Themselves to Use Tools","work_id":"9bce40c8-cfd7-4983-80e0-c3bd4402322a","shared_citers":9},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":8},{"title":"Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection","work_id":"7316de4a-d07d-41de-88ee-509f9b52e462","shared_citers":8},{"title":"Training language models to follow instructions with human feedback","work_id":"52aff42f-4fa9-4fcf-bdb3-1459b9bebf65","shared_citers":8},{"title":"WebGPT: Browser-assisted question-answering with human feedback","work_id":"e25ef3e1-4848-4cb9-bf28-67a420591165","shared_citers":8},{"title":"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models","work_id":"d1cf6693-a082-403c-ada9-dac7b96341f9","shared_citers":7},{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","work_id":"588618d7-fd41-4053-b34d-a981f8793039","shared_citers":7},{"title":"LoRA: Low-Rank Adaptation of Large Language Models","work_id":"0426219a-789e-4964-adc8-a04538510818","shared_citers":7},{"title":"Available: https://arxiv.org/abs/2401.18059","work_id":"e941569c-fee0-4d1f-bfe0-7e888d152264","shared_citers":6},{"title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","work_id":"ed240a10-5b19-406c-baa5-30803f465785","shared_citers":6},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":6},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":6},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":6},{"title":"Nikita Kitaev, Lukasz Kaiser, and Anselm Levskaya","work_id":"37c05e13-4a24-44f8-a1c4-da1bbe7223aa","shared_citers":6},{"title":"Realm: Retrieval-augmented language model pre-training","work_id":"a397ddf8-b0b7-4e32-9d59-fb6ea67ac287","shared_citers":6},{"title":"Survey of Hallucination in Natural Language Generation , volume=","work_id":"aa2fb99f-beb8-463c-8cd4-c575e96e2512","shared_citers":6},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":6}],"time_series":[{"n":1,"year":2020},{"n":2,"year":2021},{"n":3,"year":2022},{"n":1,"year":2024},{"n":65,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T08:57:52.379759+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T08:57:58.982463+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","claims":[{"claim_text":"Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, b","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"SHIELD [157] Supervised Yes DARPA-E3, ATLASv2 Yes Prec DeepSeek-R1 LogReasoner [103] Supervised Yes BGL, Spirit No F1 Qwen2.5-1.5B Data Generation, Benchmarks & Visions Karlsen et al. [71] Supervised Yes CSIC, Thunderbird, BGLYes F1 RoBERTa/GPT LogEval [22] Supervised No BGL, Thunderbird No F1 GPT-4 AD-LLM [192] Semi-Sup Yes News, Reviews, Spam No AUC LLaMA-3/GPT-4o AnomalyGen [84] Supervised Yes Hadoop, HDFS Yes F1 GPT-4o 29 6.2.2 LLM-based Approaches for Anomaly Detection.Table 4-5 summarizes ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"RAGs contribute to addressing critical limitations such as hallucinations in LLMs' responses, knowledge cutoffs, especially when LLMs are required to deal with do- main-specific questions that require precise expertise , and , importantly for the sake of transparency in LLMs ' computations, generations, and outputs , the lack of traceability [1,2]. By dynamically retrieving relevant document excerpts at inference time, RAGs ena- ble LLMs to generate responses anchored in verifiable source materi","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"reproducibility, we provide the complete data of our experiments at https://github.com/lorenzbrehme/CARE. 1 Introduction Retrieval-Augmented Generation (RAG) is an ongoing trend in research and has been continuously developed over the past years [5]. It augments their knowl- edge through the integration of external data sources and addresses hallucina- tions in large language models (LLMs) [15]. A basic RAG system consists of three components [13]: (i) Indexing component-responsible for indexing","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"\"Hallucination\", the generation of plausible but factually incorrect content, remains a major barrier to enterprise LLM adoption [12], typically categorized into intrinsic (contradicting inputs) and extrin- sic (fabricating details) types [27]. Standard mitigation strategies operate at different stages of the model lifecycle: RLHF [22] aligns models during training, while Retrieval-Augmented Generation (RAG) [15] grounds inference in external evidence. However, gen- erators may still fail due to","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Sai utilizes a Multi-Agent System (MAS) architec- ture [7, 8] integrated with a daemon-based message management system. The MAS consists of specialized agents that collaborate to decompose and execute complex tasks, supported by base LLMs for foundational reasoning and natural language pro- cessing. To incorporate domain-specific expertise, Retrieval-Augmented Generation (RAG) [9] modules are utilized to feed external knowledge directly into the agents' decision-making and code-generation proces","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Unlike passive language models, agentic systems are dynamic, autonomous agents that reason, plan, and interact with external knowledge sources [16,17]. RAG combines pre-trained parametric memory with non-parametric memory (dense vector indices of documents) for knowledge-intensive tasks, enabling models to generate more specific, diverse, and factual language than parametric-only baselines [18]. They maintain task focus, retrieve information from external resources, synthesize diverse informatio","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (31 contexts).","role_counts":[{"n":31,"context_role":"background"},{"n":3,"context_role":"method"},{"n":1,"context_role":"other"}]},"error":null,"updated_at":"2026-05-19T14:41:31.242473+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","claims":[{"claim_text":"Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, b","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T08:57:52.330486+00:00"}},"summary":{"title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","claims":[{"claim_text":"Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still limited, and hence on knowledge-intensive tasks, their performance lags behind task-specific architectures. Additionally, providing provenance for their decisions and updating their world knowledge remain open research problems. Pre-trained models with a differentiable access mechanism to explicit non-parametric memory can overcome this issue, b","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"ReAct: Synergizing Reasoning and Acting in Language Models","work_id":"407a2351-25f1-497d-b611-f77d0292a8e6","shared_citers":17},{"title":"Reflexion: Language Agents with Verbal Reinforcement Learning","work_id":"778f739e-5f55-4961-8a2a-e4736a2757f4","shared_citers":13},{"title":"Retrieval-Augmented Generation for Large Language Models: A Survey","work_id":"b80d2790-6cd9-4c87-b3c4-de404f99a80e","shared_citers":13},{"title":"AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation","work_id":"92b7eb9c-c3d8-4518-a376-06fa15dd895b","shared_citers":10},{"title":"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena","work_id":"d0c30cd7-81e1-4159-a87f-f6adca77ff08","shared_citers":10},{"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","shared_citers":9},{"title":"MemGPT: Towards LLMs as Operating Systems","work_id":"2698f5ad-c84c-40ca-b839-0912dae10ba2","shared_citers":9},{"title":"Toolformer: Language Models Can Teach Themselves to Use Tools","work_id":"9bce40c8-cfd7-4983-80e0-c3bd4402322a","shared_citers":9},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":8},{"title":"Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection","work_id":"7316de4a-d07d-41de-88ee-509f9b52e462","shared_citers":8},{"title":"Training language models to follow instructions with human feedback","work_id":"52aff42f-4fa9-4fcf-bdb3-1459b9bebf65","shared_citers":8},{"title":"WebGPT: Browser-assisted question-answering with human feedback","work_id":"e25ef3e1-4848-4cb9-bf28-67a420591165","shared_citers":8},{"title":"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models","work_id":"d1cf6693-a082-403c-ada9-dac7b96341f9","shared_citers":7},{"title":"From Local to Global: A Graph RAG Approach to Query-Focused Summarization","work_id":"588618d7-fd41-4053-b34d-a981f8793039","shared_citers":7},{"title":"LoRA: Low-Rank Adaptation of Large Language Models","work_id":"0426219a-789e-4964-adc8-a04538510818","shared_citers":7},{"title":"Available: https://arxiv.org/abs/2401.18059","work_id":"e941569c-fee0-4d1f-bfe0-7e888d152264","shared_citers":6},{"title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","work_id":"ed240a10-5b19-406c-baa5-30803f465785","shared_citers":6},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":6},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":6},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":6},{"title":"Nikita Kitaev, Lukasz Kaiser, and Anselm Levskaya","work_id":"37c05e13-4a24-44f8-a1c4-da1bbe7223aa","shared_citers":6},{"title":"Realm: Retrieval-augmented language model pre-training","work_id":"a397ddf8-b0b7-4e32-9d59-fb6ea67ac287","shared_citers":6},{"title":"Survey of Hallucination in Natural Language Generation , volume=","work_id":"aa2fb99f-beb8-463c-8cd4-c575e96e2512","shared_citers":6},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":6}],"time_series":[{"n":1,"year":2020},{"n":2,"year":2021},{"n":3,"year":2022},{"n":1,"year":2024},{"n":65,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"5cf0b0f3-cf4a-4e65-a452-37fb8a08fbb8","orcid":null,"display_name":"Aleksandra Piktus","source":"manual","import_confidence":0.72},{"id":"19b22360-f7f4-4cb9-b4fd-ab815d17f54a","orcid":null,"display_name":"Ethan Perez","source":"manual","import_confidence":0.72},{"id":"32209fff-0763-442b-a3eb-fc7c1b841804","orcid":null,"display_name":"Fabio Petroni","source":"manual","import_confidence":0.72},{"id":"63b793df-28cb-4279-a2b8-3385a04bd3d4","orcid":null,"display_name":"Naman Goyal","source":"manual","import_confidence":0.72},{"id":"ab027a6a-af2b-4f6c-8e2f-4d3c57f079b0","orcid":null,"display_name":"Patrick Lewis","source":"manual","import_confidence":0.72},{"id":"bffad31d-14ea-461f-adaf-4b6ed63eec65","orcid":null,"display_name":"Vladimir Karpukhin","source":"manual","import_confidence":0.72}]}}