5

I am a student in psychology, but I have very little familiarity with linguistics. I am doing working on flexible cognition and memory, and we are developing a task that requires participants to select objects that can be used to satisfy goals, while systematically varying the association with the object and the goal.

Goals are given as 'tasks' that can be expressed as an infinitive, i.e. "To Dig", and participants would then select a noun (from a field of distractors) that could best be used to perform the task, i.e. "Shovel" or "Stick". The latter would be an example of an object that could be used to perform the task, but is not usually associated with the task in the way that a shovel is.

We would like to have some 'objective' measure of semantic relatedness between the goal (verb/infinitive) and the object (noun). Unfortunately, Wordnet and many other online databases of semantic relatedness don't let you get measures of semantic relatedness across different parts of speech.

So, the question: Does anyone know of a toolset that can be used to extract some measure of semantic relatedness between a word pair, i.e. ('Dig', 'Stick'), across parts of speech, while simultaneously allowing me to specify the part of speech associated with each word in the pair? (to stretch the above example, I want to be able to specify that dig is to be understood as a verb, as in 'to dig' not a noun as in 'let's go to the archeological dig')

I am reasonably fluent in python, and I have already set up NLTK on my computer, but for the life of me I can't figure out how to turn it toward this problem. Hopefully the answer is there somewhere, and if so, code examples would be highly appreciated.

  • You might be interested in the notion of semantic fields. I don't know if there are any tools which use it for NLP though. – curiousdannii Dec 9 '14 at 13:46
  • I'm not very much familiar with the project: wiki.opencog.org/w/Dependency_relationship (I am only somewhat familiar with link-grammar). It looks like it might have something to offer in this respect, although it's not its primary goal. – wvxvw Dec 16 '14 at 20:49
  • 1
    WordNet similarity metrics are explained in this answer. I'd start there to get an idea of defining your semantic similarity score for lexical items with different part-of-speech tags. If you want a really generic sense of similarity, you could simply go with a word vector approach, e.g. word2vec. – user3898238 Feb 11 '15 at 3:38
1

I don't know of a toolset, but I can think of a factors that you'd need to account for to determine 'relatedness' across parts of speech:

First, consider for the eight parts of speech:

  1. verb
  2. adjective
  3. adverb
  4. noun
  5. pronoun
  6. preposition
  7. conjunction
  8. interjection

I'm taking a guess at this, but I'd say that not all eight parts of speech communicate an equal level of ' semantic relatedness'. So, evaluate how each part of speech influences the others. I'm a web developer with an interest in computational linguistics, so here's my hair-brained theory:

  • Nouns are the person, place, thing, or idea. Probably the highest level of relatedness you could evaluate is between nouns. "Shovel, dirt, hole" "John"

  • Adjectives describe nouns and pronouns. There will probably be a high level of relatedness between adjectives and nouns/pronouns. You could probably determine the level of relatedness based on proximity to the noun. "dirty shovel" "black dirt" "little John"

  • A pronoun is a word used in place of a noun. Since it's used in place of some sort of noun ("he" instead of "John"), it removes semantic relatedness. But, a verb must agree with a pronoun, so you can connect a verb to the actor with it. "He digs with the shovel" "He digs with it" instead of "John digs with the shovel"

  • Verbs tell you what a noun or pronoun is, was, or will be, or could be doing - so there's problem a very high level of 'relatedness' between nouns and verbs. So maybe verbs have equal weight as nouns? "to dig, to bury, to uncover" "John digs with the shovel" "Jane was buried with the shovel"

  • Adverbs modify verbs,adjectives, and other adverbs. So, this could get tricky. Consider the phrase, "He digs very fast with the very heavy shovel". "Very" is modifying both a verb and adjective. If your focus is on nouns, then "very" really has a very weak semantic relationship. You'd have to look for... adverbed adjectives in order to find a strong semantic relationship: "He laboriously digs with the shovel". So... maybe '-ly' adverbs have a slightly stronger relationship?

  • Preposition. A preposition communicates relationship between nouns "with the shovel, from the ground". I'd say the relationship is super low.

  • Conjunction. Conjunctions join clauses together. I'd say the relationship is very low. "He digs with the shovel and whistles"
  • interjection. Words like "ouch", "hey", or "argh" have a significant emotional relationship to something in the sentence. This could be several studies, all on their own.

In summary: I don't know of a tool, but if I were building one, it'd work like this:

  1. ignore pronouns, prepositions, conjunctions
  2. Focus on nouns and verbs
  3. Then find relatedness between adjectives and nouns
  4. And Also find relatedness between adverbs and verbs
  5. Then find relatedness between adverbs and adjectives
  6. Then find relatedness between interjections and (nouns, verbs, adverbs, adverbs modifying verbs)
| improve this answer | |

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.