Neural machine translation (NMT) looks at the entire sentence when translating, so the translation may change depending on surrounding words.
Registering a user dictionary should allow you to improve translation consistency to a certain degree. However, if too many common nouns are registered, the accuracy of the entire sentence might actually decrease, or the chance of unintentionally applying part of a registered word might increase. You will need to be careful when selecting terms.
The technology is also sensitive to input differences, handles word inconsistencies poorly, and translation results may vary greatly depending on the presence or absence of punctuation and periods. Editing the source text (pre-editing) may work well, such as removing punctuation and periods from the source text before translation.
This may also be affected by updates to the translation engine. Translation engine updates are not necessarily announced every time.
Also refer to "Translation Tips".