Something I've been pondering- spacefaring humanity makes contact with alien species who, while being intelligent enough to learn to understand human languages, their throats or vocal cords (assuming that's what they use to communicate with in the first place) aren't really suited to make the proper sounds for English (or Cantonese or Swahili, etc.) And humans have similar problems with the aliens' "native tongue"- what would be an effective way for these species to work around this problem?
If both parties can perceive the other's signals:
Receptive bilingualism, or creolization.
Han Solo and Chewbacca (Star Wars) demonstrate receptive bilingualism: each of them can understand both languages, but only speak one. So, they each speak their own language, and get along just fine.
Ryland and Rocky (Project Hail Mary) demonstrate creolization. Ryland isn't great at learning languages, and Rocky can't reproduce human speech... so they settle on a mixed language, in which Rocky pronounces Eridian words (which Ryland can learn to recognize) in English-like syntax (which Ryland already knows).
If one or both parties cannot perceive the other's signals:
Perhaps the aliens are deafblind, or at least deaf to the range of human speech. Or perhaps they can perceive our language, but they communicate in ultrasonics, or with IR light, or through biometric fields, etc.--some medium that we can't perceive. Then technological assistance is required. At minimum, you have devices which transpose each party's signals into a format which the other can perceive, and then proceed with one of the above options. Potentially, you rely on technology for automatic translation entirely.
The Written Word
There's a science fiction book (possibly by Robert L Forward) with underwater aliens with sonar, whose language closely resembles the echo-sounds of the object of the word they're saying. They developed knot-based writing to communicate over longer distances. Humans arrived, and the sound language was far too complicated and subtle to even start, but the knot system was very accessible to both.
This idea could be applied to other variations of the written word, in a shared language accessible to both, despite not being their "natural" language.
In particular, transforming rather than encrypting and decrypting ones. Each species wears a device which takes their words in either language and shifts frequency, etc. to something the other can deal with. In this model, one or more learns the language of the other, and translation is by the speaker and not a computer. The computer merely applies rules to transform the sound.
Obviously, we could get more sophisticated than that, with Google Translate But For Aliens doing live audio translations, but we're not at all lacking experience with communicating with people who can't speak.
Construct a Natural Language Model for both languages, let the AI interpret.
If you have the cooperation of both species, they each simply build a “corpus” of their language by uploading large volumes of written language. The AI doesn’t need to know what the words mean (and it never does), it simply knows how often words are used and in what circumstances.
With this built on each side, written language translations become nearly automatic. The translation only works when the language models are “parallèle” however. This means that concepts of “anger” for example, are similarly tagged in each Natural Language Model corpus. When the AI searches for terminology to say, “I’m upset”, for example, it looks in the correct place in each language model.
Now proceed to “prononciation” of each object (for us it is words and phrases, they may do things differently - AI machine learning doesn’t care)
Communication may now begin through text-to-speech and vice-versa. Here are the steps this systems would take to translate a spoken phrase on either side:
The speech to be translated is recorded and converted into a digital format, and noise is removed. It’s now “pre-processed speech audio.”
The pre-processed speech is segmented into individual objects (words or phrases for humans).
The objects/words or phrases are translated using the written NLM. Now the written data is saying the same basic thing in the other language.
Now the translated text is simply converted back into speech and played back to the listener.
That’s how modern Chatbots are converting our notes and letters into different languages more easily than even Google Translate.