This paper presents a semantic parser that scales to Freebase, trained using question-answer pairs rather than annotated logical forms. The main challenge is narrowing down the large number of possible logical predicates for a given question. To address this, the authors build a coarse mapping from phrases to predicates using a knowledge base and a large text corpus, and use a bridging operation to generate additional predicates based on neighboring predicates. On the Cai and Yates (2013) dataset, their system outperforms their state-of-the-art parser despite not using annotated logical forms. They also collect a new, more challenging dataset of question-answer pairs and improve over a natural baseline. The semantic parser uses a log-linear model over derivations, where each derivation is a tree specifying the application of combination rules. The parser constructs a distribution over possible derivations for an utterance, and uses features to guide the selection of the most likely derivation. The system includes alignment features, bridging features, and composition features to improve performance. The authors evaluate their system on two datasets: Cai and Yates (2013) and a new dataset called WebQuestions. On Cai and Yates, their system achieves 62% accuracy, outperforming the 59% accuracy of the state-of-the-art system. On WebQuestions, their system achieves 31.4% accuracy, a 4.5% improvement over a natural baseline. The system is able to handle a wide range of questions, including those involving rare predicates and complex logical forms. The paper also discusses the challenges of semantic parsing, including the need for a manageable set of predicates and the importance of features in guiding the parser. The authors propose a new bridging operation that generates predicates based on neighboring predicates, which is crucial for handling weakly or implicitly expressed predicates. The system is able to handle a wide range of questions, including those involving rare predicates and complex logical forms. The authors conclude that their system provides a significant improvement over previous approaches and is able to handle a wide range of questions on Freebase.This paper presents a semantic parser that scales to Freebase, trained using question-answer pairs rather than annotated logical forms. The main challenge is narrowing down the large number of possible logical predicates for a given question. To address this, the authors build a coarse mapping from phrases to predicates using a knowledge base and a large text corpus, and use a bridging operation to generate additional predicates based on neighboring predicates. On the Cai and Yates (2013) dataset, their system outperforms their state-of-the-art parser despite not using annotated logical forms. They also collect a new, more challenging dataset of question-answer pairs and improve over a natural baseline. The semantic parser uses a log-linear model over derivations, where each derivation is a tree specifying the application of combination rules. The parser constructs a distribution over possible derivations for an utterance, and uses features to guide the selection of the most likely derivation. The system includes alignment features, bridging features, and composition features to improve performance. The authors evaluate their system on two datasets: Cai and Yates (2013) and a new dataset called WebQuestions. On Cai and Yates, their system achieves 62% accuracy, outperforming the 59% accuracy of the state-of-the-art system. On WebQuestions, their system achieves 31.4% accuracy, a 4.5% improvement over a natural baseline. The system is able to handle a wide range of questions, including those involving rare predicates and complex logical forms. The paper also discusses the challenges of semantic parsing, including the need for a manageable set of predicates and the importance of features in guiding the parser. The authors propose a new bridging operation that generates predicates based on neighboring predicates, which is crucial for handling weakly or implicitly expressed predicates. The system is able to handle a wide range of questions, including those involving rare predicates and complex logical forms. The authors conclude that their system provides a significant improvement over previous approaches and is able to handle a wide range of questions on Freebase.