The study integrates the use of information alignment algorithm and DR-Reformer network to construct the DR-Reformer English translation model based on optimal transportation. The information alignment algorithm realizes the modeling of correspondence between source and extended vocabularies and can output high-quality aligned utterances, while the DR-Reformer network can complete the tasks of vocabulary long-distance encoding and memory allocation, which improves the computational efficiency and scalability of the model. Under the constructed experimental environment, the cultural adaptability of the model in English translation of intangible heritage in folk culture tourism is evaluated according to BLEU and TER indicators. The G-mean value of the model is 0.912 when the feature dimension of the test dataset is 100 and the positive: negative ratio is 1:2, at which time the recognition of intangible heritage information sentences is the best. The model in this paper has a stable BLEU value of 37.5 and a stable TER value of 57.49 for the translation of sentences containing intangible heritage information, which has a better translation effect compared with the comparison model. The accuracy, fluency and logic of the intangible heritage texts translated by the model meet the reading requirements of tourists, with an average score between 7.77 and 8.34, and have better cultural adaptability.
