This paper presents a systematic comparison between context-counting and context-predicting semantic vectors, a new approach in distributional semantics. The authors, Marco Baroni, Georgiana Dinu, and Germán Kruszewski, from the Center for Mind/Brain Sciences at the University of Trento, Italy, evaluate these models on a wide range of lexical semantics tasks and across various parameter settings. Despite the initial skepticism surrounding context-predicting models, the results show that they outperform classic count-based approaches in most tasks, achieving state-of-the-art performance in several cases. The study highlights the robustness and effectiveness of context-predicting models, which are trained using supervised learning to maximize the probability of observed contexts, rather than relying on heuristics. The authors also discuss the limitations and potential future directions for both types of models, suggesting that context-predicting models have significant advantages in certain applications, such as phrase representation and fusion of language and vision.This paper presents a systematic comparison between context-counting and context-predicting semantic vectors, a new approach in distributional semantics. The authors, Marco Baroni, Georgiana Dinu, and Germán Kruszewski, from the Center for Mind/Brain Sciences at the University of Trento, Italy, evaluate these models on a wide range of lexical semantics tasks and across various parameter settings. Despite the initial skepticism surrounding context-predicting models, the results show that they outperform classic count-based approaches in most tasks, achieving state-of-the-art performance in several cases. The study highlights the robustness and effectiveness of context-predicting models, which are trained using supervised learning to maximize the probability of observed contexts, rather than relying on heuristics. The authors also discuss the limitations and potential future directions for both types of models, suggesting that context-predicting models have significant advantages in certain applications, such as phrase representation and fusion of language and vision.