Thanks for the feedback, @Elisardojm.
Regarding the translation of specific words (such as "una" or "para") in the examples, we cannot do much about it directly. Content translation integrates existing machine translation services, and improving these services is part of their corresponding teams work. What we do is exposing the user corrections made for them these teams to process and make improvements based on them. So improving those initial translations for each article you translate, you are already helping the translation services to improve in the future.
Regarding the translation of templates there are some aspects that help Content translation to do a better job. Templates exist independently in each wiki, so Content translation relies on metadata to find the equivalent template first, and then find the equivalent parameters to copy over the data.
For example, Content Translation finds that "Cita novas" is the equivalent template to "Cita noticia" because they are connected in Wikidata. However, the tool cannot find the equivalent parameters between the two templates because of the lack of metadata. For parameters with the same name in both templates (such as "autor") the tool can copy the information, but there is no information telling the tool that "fechaacceso" and "dataacceso" are equivalent. In order to support this, editors need to provide the corresponding templateData. In particular, the "dataacceso" of the Galician template needs an alias named "fechaacceso" for Content translation to know they are equivalent.
The underlying problem in this case is that templates were not designed to work well across languages, so we rely on metadata to do the best possible mapping.
Hope this is helpful.