I want to do a very simple job: given a string containing pronouns, I want to resolve them.
For example, I want to turn the sentence "Mary has a little lamb. She is cute." into "Mary has a little lamb. Mary is cute.".
I use jave and Stanford Coreference which a part of Stanford CORENLP. I have managed to write some of the code but I am unable to complete the code and finish the job. Below is some of the code which I have used. Any help and advice will be appreciated.
String file=" Mary has a little lamb. She is cute.";
Properties props = new Properties();
props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref");
StanfordCoreNLP pipeline = new StanfordCoreNLP(props);
Annotation document = new Annotation(file);
pipeline.annotate(document);
List tokens = new ArrayList();
List sentences = document.get(CoreAnnotations.SentencesAnnotation.class);
for(CoreMap sentence: sentences)
{
Map graph = document.get(CorefChainAnnotation.class);
System.out.println(graph);
for (Map.Entry entry : graph.entrySet()) {
CorefChain c = (CorefChain) entry.getValue();
CorefMention cm = c.getRepresentativeMention();
System.out.println(c);
System.out.println(cm);
}
}
Pronoun resolution is not an easy job. Ideas like the closest matching entity are often used. But there are exceptions to all these. What is your approach to pronoun resolution?
What do you mean by "What is your approach to pronoun resolution", could you give me more explanation please.
Hi,
Please see this link. It contains a comprehensive example of getting co reference resolution
http://blog.pengyifan.com/resolve-coreference-using-stanford-corenlp/
I hope this is useful for you.
There is a lot of literature about pronoun resolution. For example, Gerlof Bouma wrote his PhD thesis on pronoun resolution using Optimality Theory. His PhD thesis can be found on the Internet, but you can also get a small taste here (optimality theory applied to construct an algorithm for pronoun resolution in Dutch):
http://www.let.rug.nl/~zwart/college/docs/minor/bouma.pdf
A plausible semantic solution to your problem for the concerned example could be as follows:
Keep 3 bounded queues for each type of subjects: {male, female, objects}. Have an extensible mapping knowledge base to identify the which category a given subject fall under. E.g, names like Mary fall under female category, names like Mark fall under male, the rest as objects (places, animals, and other things).
You may keep the queue size for each category to be two. E.g, you could say "Mark was talking to Anthony. He told him that..." in here 'he' would refer to Mark and 'him' would refer to Anthony. For simplicity you may keep the queue size as one considering the simple case you have illustrated in the example.
Each time a pronoun is referenced, replace it accordingly with the names in the respective queues.
Each time a new subject is encountered, queue in the subject to the queue that matches the respective category (one of male, female or object).
The above solution would certainly work for the simple cases as the one described in the question. Accuracy for complex forms may vary. Complex forms would require more use of deductive systems. You may also use Turing Machines to process such tasks.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication