Information on searching for semantic relationships
This application searches for semantic relationships in a text by analysing the statistical properties of words.
It is not based on rules but on the probability that two words can appear in the same phrase without having a relationship.
It creates an n-dimensional representation of words (PLSA) by using the statistical properties of the words which appear next to them as coordinates. This demo uses European Parliament debates as its corpus.
Why have we developed this?
This technology is an integral part of a more complicated project able to extract translated terminology from the web.
For example, if you want to find on the web the English translation of ‘Metallizzazione’, it will be difficult to find bilingual sites from where the information can be extracted. But you will find on google more than 50.000 Italian pages which talk about "Metallizzazione". From these pages, you will discover that ‘Metallazzione’ has semantic relationships with "vuoto", "impianto", "vernice", "finitura", "metallo" for which the English translations can be easily found. At this point, you can search for what the words "vacuum", "plant", "paint", "metal" have in common and the answer will be "Metallization", the translation you were looking for!.
I want it!
If you are interested in this technology, please read more on Translated Labs and our services for natural language processing.
Explore our experiments
The Language Identifier automatically detects the language of a written text. It can also be used to identify the topic of a written text in a language you do not understand.Learn more
What do the words airplane, bird, and helicopter have in common? This application searches for semantic relationships in a text by analyzing the statistical properties of words.Learn more
What happens when you translate an English sentence into Japanese, and then again into English, as if it was an infinite loop? Well, give it a try! And don't forget to share the funniest results with your friends.Learn more