This paper presents a semantic parser that scales to the Freebase knowledge base, focusing on mapping natural language questions to answers via latent logical forms. The main challenge is narrowing down the vast number of possible logical predicates for a given question. The authors address this by using a coarse alignment based on Freebase and a text corpus, and a bridging operation that generates additional predicates based on neighboring predicates. On the Cai and Yates dataset, their system outperforms the state-of-the-art parser without annotated logical forms. Additionally, they collect a more realistic and challenging dataset of question-answer pairs, achieving 31.4% accuracy, a 4.5% improvement over a natural baseline. The paper also discusses the construction of a lexicon for mapping phrases to predicates, the use of POS tags and denotation features for composition, and detailed experimental results on both datasets.This paper presents a semantic parser that scales to the Freebase knowledge base, focusing on mapping natural language questions to answers via latent logical forms. The main challenge is narrowing down the vast number of possible logical predicates for a given question. The authors address this by using a coarse alignment based on Freebase and a text corpus, and a bridging operation that generates additional predicates based on neighboring predicates. On the Cai and Yates dataset, their system outperforms the state-of-the-art parser without annotated logical forms. Additionally, they collect a more realistic and challenging dataset of question-answer pairs, achieving 31.4% accuracy, a 4.5% improvement over a natural baseline. The paper also discusses the construction of a lexicon for mapping phrases to predicates, the use of POS tags and denotation features for composition, and detailed experimental results on both datasets.