madoctopus
Supreme Member
- Apr 4, 2010
- 1,388
- 3,709
You can NOT build an automated spinner with just grammar checking. TBS already does grammar checking. What you need is more advanced NLP. Even so, it is stil imposible to get it to work well because of the word senses and context. For example you can have:
Once your computer boots and you play The Puss in Boots, you realize it is not a movie about boots but about a pussy.
Correct and makes sense. Do you think there's a chance to spin that and maintain the message correctly?
Hence, Impossible to accurately detect sense. You can detect the POS (part of speech) but that is not enough. Even POS tagging is not entirely accurate without custom training on a specific domain/corpora.
That being said, you can use more advanced NLP to improve the results of automated spinning a bit, but you still won't take it to a readable level. I do know at least two people who managed to successfully build automated spinners that produce very good results. However they have limitations - they either don't produce completely readable text or they require a certain amount of human intervention and work well only for articles with sentences structured in a certain way and no figures of speech or unusual expressions. Hence, they work well for spinning articles you write from scratch but not a solution to 100% auto generate content by spinning all the news/articles/blog posts that were created on the internet in a particular day. By the way, that would be my wet dream in terms of SEO - to be able to rewrite all content produced on the internet in a day, every day.
So what can you do? It depends what you want. It is quite easy to build a content generator that can not trigger flags with Google. All you have to do is:
The other solution, for those who do not have the knowledge and skills to build a complex generator as I did, is to simply hand spin lots of articles using multi-level spinning. Frankly, considering that you don't need any particular skills (aside from spinning) this is a very doable solution.
- The shoelaces from my boots are dirty. - boot = shoe. It is a noun.
- The computer boots quite fast. - boot = start - a process a computer does at startup. It is a verb.
- Jack gave him the boot with a satisfied grin on his face. - boot is part of an expression - "to give the boot" - to send him away. Here it is a complement but that is useless information because it is part of an expression.
- The Puss in Boots had a great success. - Here is a proper noun, it is part of the name of a movie.
Once your computer boots and you play The Puss in Boots, you realize it is not a movie about boots but about a pussy.
Correct and makes sense. Do you think there's a chance to spin that and maintain the message correctly?
Hence, Impossible to accurately detect sense. You can detect the POS (part of speech) but that is not enough. Even POS tagging is not entirely accurate without custom training on a specific domain/corpora.
That being said, you can use more advanced NLP to improve the results of automated spinning a bit, but you still won't take it to a readable level. I do know at least two people who managed to successfully build automated spinners that produce very good results. However they have limitations - they either don't produce completely readable text or they require a certain amount of human intervention and work well only for articles with sentences structured in a certain way and no figures of speech or unusual expressions. Hence, they work well for spinning articles you write from scratch but not a solution to 100% auto generate content by spinning all the news/articles/blog posts that were created on the internet in a particular day. By the way, that would be my wet dream in terms of SEO - to be able to rewrite all content produced on the internet in a day, every day.
So what can you do? It depends what you want. It is quite easy to build a content generator that can not trigger flags with Google. All you have to do is:
- Have mostly correct grammar
- Have mostly common n-grams - in other words NOT have n-grams that do not exist. For example a sentence like "the car eat an apple". "car eat an apple" has tiny/no chances to exist in real text because a car does not eat fruit. This is one of the reasons why Markov chains do not work well.
The other solution, for those who do not have the knowledge and skills to build a complex generator as I did, is to simply hand spin lots of articles using multi-level spinning. Frankly, considering that you don't need any particular skills (aside from spinning) this is a very doable solution.
