This article introduces an automated English-writing reviser (AI-Reviser) based on advanced AI/ML technology. It was developed during the summer of 2017 using my spare time, and has gradually gained popularity (reached more than 78,000 page views in around 6 months) especially among students whose native language is not English.
Table of Contents:
- Product design
- Neural-machine-translation loops
- Formatted output
- Evaluating the performance (Comparing it to other available tools)
There is a big market in revising English writings, since 95% (>6,500,000,000) of the population on this planet are not native English speakers. Assuming 5% of them need to write in English the volume of the potential market is 325,000,000 users…
As an international student in the United States, I need to communicate with my adviser by emails almost everyday, and write reports and articles very often. There is a strong demand in correcting grammar errors, and improper expressions.
I’ve checked most of the available tools online, but found none of them performs sufficiently well: most of them can distinguish the conventional grammar errors, but almost none of them can tell (or provide suggestions) if your way of expression sounds native.
This motivates me to develop an English reviser based on the cutting-edge machine learning technology.
2. Product design
I found most of the available tools have a very complicated user interface: when going to their websites you are either exposed to much irrelevant information or required to login, download, or pay before you can use their revising tool. This is quite annoying for those who would like to save not just time and money, but also “attention”.
Considering the weakness of the existing tools, I designed the UI by myself following the principle of “Laziness” (or call it Least Action? if you prefer), and set such a list of requirements for AI-Reviser:
- You can find the inputting box and button right after you open the website;
- No boring grammar analysis—instead, it provides the corrected results to you directly for your consideration;
- Not just correcting the grammar errors, but also improper way of expressions (i.e., rephrasing your expressions to make them sounding more native);
- The technology used must be the most advanced.
Here is how the resulting UI looks like (In Chinese, for the English version, go directly to the official website to have a look):
A good modifier must be robust enough to accommodate the confusing formats of English beginners’ inputs. I found a common problem for Chinese children is that they often omit the blank space behind the period. This seemingly insignificant issue could bring trouble to machine learning. After searching the literature, and trying out the open source software package, I didn’t find a good solution for that, and was forced to code for it myself. After some thought, I used several lines of code to judge if the user often ignores the space after a period, the problem is magically solved. – Since then, I no longer have to worry about the tool would be broken by those naughty children.
4. NMT Loops
The full name of NMT is Neural Machine Translation. The famous translators such as the latest version of Google Translator, and the Colorful-cloud Translator are all using a similar technology. The feature of this technique is that it does not require semantic analysis and can feed a sentence directly to the neural network for training. AI-Reviser also follows this technique.
Traditional grammar modifiers are like an English teacher with a Chinese nationality who will teach you a lot about grammars. In contrast, AI-Reviser works like a native English speaker who has never learned the detailed grammar rules. She relies entirely on her “sense of language” — The feedback she gives is always something like “Emm.., this place reads awkward. How about changing it to be like this?” Because of this, AI-Reviser has an unmatched ability to check if an expression sounds native.
In order to enhance the modification effect, I requires the reviser to loop around the core modification network until the convergent condition is met. Our tests show that the convergence is quite fast.
5. Formatted output
The output also adheres to the principle of “laziness”. It only provides the two most useful versions of output: The first version is the revised text, easy to copy and paste; the second version is sentence-by-sentence comparison, easy to analyze. The Jaccard Index is used behind this intelligent association of sentences before and after modification. In addition, the modified places will be automatically marked with color, so that the user can quickly see where the changes are.
6. Evaluating the performance (Comparing it to other available tools)
By comparing the performance of AI-Reviser to that of several most popular existing English-revising tools, I found that AI-Reviser outperforms almost all of them in many aspects. Here is an example showing that AI-Reviser provides the best corrections to a sentence randomly chosen:
You can use the following text to make a more thorough test (Please remember to register and login, otherwise, it will only revise the first 3 sentences):
English is a internationaly language which becomes importantly for modern world.
In China, English is took to be a foreigh language which many student choosed to learn. They begin to studying English at a early age. They use at least one hour to learn English knowledges a day. Even kids in kindergarten have begun learning simple words. That’s a good phenomenan, for English is essential nowadays.
In addition to, some people think English is superior than Chinese. In me opinion, though English is for great significance, but English is after all a foreign language. it is hard for people to see eye to eye. English do help us read English original works, but Chinese helps us learn a true China. Only by characters Chinese literature can send off its brilliance. Learning a country’s culture, especial its classic culture, the first thing is learn its language. Because of we are Chinese, why do we give up our mother tongue and learn our own culture through a foreign language?
Official Website: https://www.aireviser.com