The utopia of AI. Applying natural language processing techniques

Summary

NLP is living its golden age, in which the disruption of pre-trained language models and transfer learning have been able to dramatically improve the performance of some NLP tasks exceed human-level accuracy. However, what happens when the understanding of language involves speech and its transcription has poor semantic signals? In these cases, errors are propagated throughout the entire methodological life cycle, from the annotation to the final solution, complicating the process so much that AI systems appears to be a huge challenge or maybe even a utopia?