This paper introduces a simple unsupervised learning algorithm called PMI-IR for recognizing synonyms. PMI-IR uses Pointwise Mutual Information (PMI) and Information Retrieval (IR) to measure the similarity of word pairs by querying a web search engine. The algorithm is evaluated using 80 TOEFL synonym test questions and 50 ESL synonym test questions, achieving a score of 74%. In comparison, Latent Semantic Analysis (LSA), another unsupervised learning algorithm, achieves a score of 64% on the same TOEFL questions. The paper discusses the potential applications of PMI-IR and the implications of the results for LSA and Latent Semantic Indexing (LSI). The author suggests that PMI-IR's performance is due to its use of a larger data source and a smaller chunk size compared to LSA. The paper also explores the relationship between PMI-IR and query expansion techniques in Information Retrieval systems. Finally, the author discusses the potential applications of PMI-IR in lexical database construction, IR systems, and automatic keyword extraction.This paper introduces a simple unsupervised learning algorithm called PMI-IR for recognizing synonyms. PMI-IR uses Pointwise Mutual Information (PMI) and Information Retrieval (IR) to measure the similarity of word pairs by querying a web search engine. The algorithm is evaluated using 80 TOEFL synonym test questions and 50 ESL synonym test questions, achieving a score of 74%. In comparison, Latent Semantic Analysis (LSA), another unsupervised learning algorithm, achieves a score of 64% on the same TOEFL questions. The paper discusses the potential applications of PMI-IR and the implications of the results for LSA and Latent Semantic Indexing (LSI). The author suggests that PMI-IR's performance is due to its use of a larger data source and a smaller chunk size compared to LSA. The paper also explores the relationship between PMI-IR and query expansion techniques in Information Retrieval systems. Finally, the author discusses the potential applications of PMI-IR in lexical database construction, IR systems, and automatic keyword extraction.