Mike Lewis · page 2

29 papers in our library · 2014–2024 · page 2 of 2

Papers

The most cited first (OpenAlex and our library).

Contrastive Decoding Improves Reasoning in Large Language Models (2023 · 68 citations)
LLMs are usually decoded differently depending on the task: open-ended writing relies on truncated sampling, while reasoning tasks typically use greedy decoding.
Questions Are All You Need to Train a Dense Passage Retriever (2023 · 65 citations)
Dense passage retrievers for open-domain QA usually depend on large labeled sets of question–passage pairs, plus extra machinery such as hard-negative mining and cleaning of positive examples.
In-context Pretraining: Language Modeling Beyond Document Boundaries (2023 · 61 citations)
Standard LM pretraining packs randomly chosen documents into each context window. Earlier documents therefore tell the model nothing about later ones, so the model gets no practice using information across document boundaries. The authors (Meta AI, University of Washington, AI2) propose In-Context Pretraining. It…
Improved CCG Parsing with Semi-supervised Supertagging (2014 · 58 citations)
CCG parsers depend heavily on supertagging, which assigns each word a rich lexical category. This step suffers from sparse labelled data, unseen words, and errors passed on from a separate POS-tagger, especially outside newswire. Lewis and Steedman replace the POS-tagging and supertagging stages of the C&C pipeline…