Expectation–maximization (EM) is a critical and very accurate machine translation method used in the Bangla–Odia system. Here, the entire mathematical equation is computed and displayed using a corpus of Bangla– Odia words. To determine the probability value with the highest degree of likelihood, the “argmax” function, which examines the one-to-one, one-tomany, and many-to-many mappings between two or more words in source and target language phrases, is used in conjunction with this EM technique. A mathematical formula can be used to determine the lexical connections between the terms of the parallel sentences to show which word in the target language is congruent with which word in the source language. Because the EM technique loops, it is simple to compute specific probability values in terms of maximum likelihood estimation (MLE) and then iteratively determine the word associations between the source and destination languages. Because 416machine translation is currently one of the main focuses of natural language processing (NLP) research work with some practical obstacles, determining the MLE or maximal a posterior (MAP) of the parameters in the probability model is a very challenging task. The purpose of this study is to describe the different lexical issues that can arise when reading bilingual materials that were translated from Bangla into Odia. The way the EM algorithm and MLE method handle the “word divergence” or “lexical divergence” problem is one of the fundamental aspects of the MT system. The MLE technique is employed in part to enhance the probability values and provide more precision, even if the EM algorithm is used to solve it. Bilingual dictionaries or lexical databases are employed in this situation because they are typically used at the phrase level to address word divergence or lexical divergence difficulties. The biggest challenge is in identifying single-word units from the source text that are transformed into multiword units in the target text. The findings of the trials leverage the bilingual Bangla–Odia corpus, which was produced using the EM algorithm, to pinpoint a productive combination for accurately resolving the lexical divergence problem.