6

I'm looking for something like the (really excellent and useful) MRC database that includes a measure of semantic relatedness for a given pair or set of words in colloquial American English.

I've found quite a few papers on how to create a measure of semantic relatedness, but that's not what I'm looking for at the moment. There's also the very promising Measures of Semantic Relatedness Server at RPI's Cogworks Lab, but it appears to be down and may be vestigial.

Does this exist?

  • 1
    Have you seen Wordnet and Framenet? – jlawler Jul 25 '13 at 17:03
  • No, I haven't, and they're both good things to know about. But neither seems to do what I need (produce a rating of relatedness between pairs or sets of words), unless I'm missing it: they both classify each word as in or out of a given set, but I can't find any details on how central a word is to that set (which would be necessary for assessing relatedness on a more fine-grained scale). – Krysta Jul 25 '13 at 17:22
  • You can't get centrality until you can demonstrate membership, and that's not simple, especially when dealing with metaphors. And you can't have centrality at all without some metric to measure convexity, and that, too, is likely to be context-sensitive. – jlawler Jul 25 '13 at 17:26
  • Did you ever find an online database to measure semantic relatedness between pairs of words? I am struggling with the same thing. I was using EAT website but it is now crashed and I cannot access it and I am back to square one. – user17111 May 18 '17 at 11:05
2

It's not an online resource, but NLTK provides several ways to get measures of semantic similarity:

2

I will suggest two fields that might be of interest, although I can't say for sure they will give you what you need.

  • Ontologies organise lexical items hierarchically.

Wikipedia has an article on (web) ontologies, and there is also a list of ontology tools. If you find a ready-made ontology for English and a way of querying the number of nodes one has to traverse to travel from word 1 to word 2 this would provide a measure of semantic distance. I quickly skimmed the list of tools and haven't found anything ready-made that can achieve this. Perhaps another Stack Exchange (Stack Overflow?) would be a good place to ask for that.

  • Distributional semantics is based on the assumption that words that frequently occur in similar contexts are closely related in meaning.

Given a large enough corpus, distributional-similarity data can be used to compute semantic relatedness. I suppose the ideal solution for you would be a table with all words as rows and columns, giving semantic distance scores for each pair of words.

My brief search didn't turn up anything like that, but you may be able to build your own semantic distance matrix by acquiring a large corpus and using the tools and guides provided by the website for a course on distributional semantics at the European Summer School on Logic, Language and Information 2009. Also, people working in the COMPOSES project might already be able to share something like a distance matrix with you or point out other sources.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.