This paper presents and compares various single-word based alignment models for statistical machine translation, focusing on the IBM alignment models (IBM-1 to IBM-5) and the Hidden-Markov alignment model (HMM). The authors discuss the impact of different alignment models, smoothing techniques, and modifications on the quality of alignments. They propose using the quality of the Viterbi alignment compared to a manually produced reference alignment as an evaluation criterion. The study shows that models with first-order dependence and a fertility model outperform simpler models like IBM-1 and IBM-2, which lack these features. The paper also explores methods to combine alignments from both translation directions to improve precision and recall. Experimental results on the VERBMOBIL and HANSARDS tasks demonstrate the effectiveness of the proposed models and techniques.This paper presents and compares various single-word based alignment models for statistical machine translation, focusing on the IBM alignment models (IBM-1 to IBM-5) and the Hidden-Markov alignment model (HMM). The authors discuss the impact of different alignment models, smoothing techniques, and modifications on the quality of alignments. They propose using the quality of the Viterbi alignment compared to a manually produced reference alignment as an evaluation criterion. The study shows that models with first-order dependence and a fertility model outperform simpler models like IBM-1 and IBM-2, which lack these features. The paper also explores methods to combine alignments from both translation directions to improve precision and recall. Experimental results on the VERBMOBIL and HANSARDS tasks demonstrate the effectiveness of the proposed models and techniques.