Text Compactor | The Evolution of a Text Summarizer
[yop_poll id=”2″]
According to their website, text compactor is a website that was created in 2010 as a free online automatic text summarizer. It was based on an open text summarizer software that is now defunct. The summarization tool worked by determining the frequency of words within sentences and giving each sentence a score. The sentence(s) with the highest scores were deemed the most relevant and used for the summary. It was recommended only for non-fiction and aimed to save busy students, teachers, and professionals time by quickly sifting through large content and papers.
Text Compactor | The Evolution of the Text Summarizer
Fast forward to today and summarization and paraphrasing tools like text compactor show their age. Older tools like text compactor were cutting edge in their time and offered a unique look into what was and still to come. Text summarization has moved far beyond counting words in sentences and scoring them.
Open AI GPT-3 Text Summarizer
Enter Open AI, an artificial intelligence company founded in San Francisco, CA in December 2015. The company has trained its AI model to recursively process and summarize books of any length. They use natural language processing with a fine-tuned model from GPT-3.
According to an Open AI blog post, the model is capable of generating sensible summaries for entire books. It produces a 6/7 rating 5% of the time, which is comparable to a summary written by a human who has read the book several times. The AI model can achieve a 5/7 rating 15% of the time. According to their post and using Alice’s Adventures in Wonderland as an example, the summarization tool works as follows:
- Alice’s Adventures in Wonderland is processed – 26,449 words.
- The original text is divided into sections and each section is summarized – 66 summaries, 6,024 words
- Section summaries are summarized again into higher-level summaries – 6 summaries, 830 words
- The summarizing process is repeated until a complete summary is achieved – complete, single summary at 136 words
The model uses recursive task decomposition. This is the breaking down of summaries into smaller sections so humans can more easily review each summary for accuracy. It also allows each part of the summary to more easily be tied to the section of the book it came from. Lastly, it allows books of unbounded length to be summarized and isn’t constricted by the length of the transformer model used.
Achieving Perfect Text Summaries
Artificial intelligence for generalized text processing such as summarization have taken a huge step forward. AI in just 10 years has went from counting words (like Text Compactor did) to intelligent models that are capable of teaching themselves.
However, we’re still a long way from the end game. As noted above by the Open AI team, their latest summarization model can achieve a 6/7 rating 5% of the time. Although this is fantastic, it is only 5% of the time.
Just like Text Compactor giving us a glimpse of the future in 2010, the Open AI GPT-3 model is our current lens. In the future, we won’t be talking about 5/7 scores 5% of the time, but 7/7 scores 100% of the time. It’s only a matter of time before artificial intelligence not only impresses with what it’s capable of doing, but will become highly accurate, dependable, and expected from us as part of our daily lives.
From Text Compactor to Wordbot to Beyond
Test compactor made a valiant effort at summarizing text, but is from a past generation of tools. Our wordbot tool uses the above GPT-3 summarization technology and is a member of the current generation. You can visit worbot.io to use wordbot’s text summarizer. Who knows what the next generation of text summarizing tools will bring, but we sure can’t wait to find out. These are exciting times, so stay tuned.
