Text paraphrasing is a challenging task. You have to deal with grammar, lexical forms, words meanings and changing sentences often results into changing the meaning of the original sentence too.
That's why we have been working hard at Webit to finally release a model that eases efforts of developers and final users by providing text paraphrases at different grades of complexity.
Let's give a look to an example input:
In ancient times Catania was associated with the legend of Amphinomos and Anapias, who, on occasion of a great eruption of Etna, abandoned all their property and carried off their aged parents on their shoulders. The stream of lava itself was said to have parted, and flowed aside so as not to harm them.
And here is the output:
In ancient times, Catania was associated with the legends of Amfinomas and Anapias, who gave up all their possessions and carried their elderly parents on their shoulders during the Etna outbreak. The lava flow itself is said to have broken up and flowed to one side so as not to hurt them.
This is very close to the expected behaviour. We achieved this result using very_high entropy and no salt. At the cost of some very few inaccuracies, we got an original new sentence with flipped form and words that still mantain the original meaning.
But applying a very_high entropy is also useful for changing shorter sentences, which are known to be more complicated to be modified without meaning alteration or loss. Let's see another example:
The world is bigger than just America and Canada
And here is the output:
The world is not just US and Canada
"Bigger than" and "not" represent a very different meaning when taken apart, but such a replacement still makes the sentence meaning intact while still being successful at rewriting the phrase. We achieved this output using very_high entropy and no salt.
"But let's get a little more deeper into the accepted parameters for the paraphrase endpoint.
Entropy is the term we use to identify the grade of shuffling and fantasy provided by our AI into rewriting your sentence. The higher the entropy, the more different the final phrase is. But shuffling things too much may result in unexpected or altered behaviour, so that's why we provided 5 different grades of entropy, to allow developers and final users to do some fine tuning up to the expected result.
Salt is a completely different type of processing. We use such term to define the words as "pieces of salt" placed up on the rewritten text, by replacing nouns and adjectives or verbs with synonyms and equivalent forms. The salt can be an integer range from 0 to 100, allowing for a more detailed tuning. We often do without salt because entropy is just very good at shuffling things enough!
Start using paraphrase
Our website has plenty of live testdrive examples available without having to sign up, but paraphrase requires more complex computation that has to pass under a controlled flow of requests. But don't worry, since you can still try it live using our montlhy offered free quota and start playing with it at the located at our RapidAPI playground.