- Part of Speech tagging! Annoyed by having to translate poems into parts of speech? Me too. im3 will give it go at doing it for you. Accuracy-wise, it is hovering at about 85-90%. But it's a learning AI, so the more times you use the training module and correct its mistakes, the better im3 will get at tagging parts of speech. This will make generating from a template a million times faster and better.
- A scripting language! im3 comes with SMUP scripting language which allows you to save generated text to variables and call them redundantly. It allows you to make function calls and generate words from parts of speech, pronunciation keys, or the number of syllables. This will make generating metered verse possible.
- Vastly improved dictionary interface. The new dictionary interface operates off a master dictionary which will allow you to copy words in bulk, & search by text, part of speech, or phoneme. You can of course load and save dictionaries, and set which dictionary a template generates from. The master dictionary contains around 130,000 words and can hold as many as 200,000.
- Improved n-gram biasing. It will be much easier to load and save n-gram biases due to improvements on the way IM saves n-gram information. Load times have been vastly reduced and optimizations have been made to the EAT function which makes eating in text less time consuming.
- im3 still includes an option to speak back the poem in a computer voice, but this time, you will be able to control the rate at which the computer reads back the poem.
- HTML docs included and can be referenced from im3's GUI.
- Video tutorials by yours truly will help shorten the learning curve.
- Semantic tagging. This will be introduced in later updates, but for right now, you can make your own dictionaries any way you please.
- im3 uses Carnegie Melon's pronouncing dictionary, which is superior to im2's.
Release should be live within the next couple of days. This is a pleasant demonstration of a very early incarnation of im3. Enjoy, & be well.