Author(s): Ananya Banerjee Natural Language Processing Unraveling Coreference Resolution in NLP! Coreference Resolution Coreference resolution is an essential task in Natural Language Processing (NLP). Before we can understand Coreference Resolution it’s important to first understand what discourse is. A sequence of sentences that occur in a context of NLP is called discourse. Discourse will naturally include references to entities and entities being discussed. These references are referred to by the term “mention”. Ana, a graduate student at UT Dallas is an example of a discourse. Her favorite part of the institute is Natural Language Processing. She enjoys blogging, singing and dancing. These entities could be “Ana”, Natural Language Processing, or “UT Dallas”. The entities “Ana”, “The Institute” and”She” refer to “Ana”, while “Her” refers to “UT Dallas”. Reference, as used in NLP is the linguistic process in which one word of a sentence, discourse, or paragraph may be referring to another word, entity, or both. Referencing is the task of solving such references. The above examples show two types of Reference Resolution: “She” and “Her” refer to entity Ana and “the Institute” refers to entity UT Dallas. We can summarize: Reference Resolution is the task of solving such references. Discourse is any sequence of sentences that occurs one after another in NLP context. Coreference Resolution is the procedure of solving pronouns in order to determine which entities they are referring to. This is also known as Reference Resolution. It can be any entity, such as a person or place. Referent refers to the object being referred. In the example above, the referent is “Ana”. Referring expressions are the mentions and linguistic expressions in the discourse. Corefer [1]. is a combination of referring expressions that all refer to the same discourse entity. Let’s take a look at another example in order to better understand the concept. Exemple Discourse: Elon Musk was born June 1971.. SpaceX’s founder, CEO, chief engineer, and designer, he is also the most well-known. He is well-known as the brain behind Neuralink. Referring Expressions Elon Musk He, The 49 Year old. Corefering Expressions Elon Musk Elon Musk: Elon Musk He, Elon Musk The 49year old. Now that you have an understanding of coreference resolution it is important to know what type of references might exist in text. Knowing the type of references can help us to devise strategies for resolving them. There are two types of references: Endophor and Exaphor. Endophor is an entity which appears in the discourse. Exaphor is an entity which does not appear within the discourse. Endophor Sentence Example: Ana loves reading. She just read a beautiful story. Exaphor sentence: Example: Pick that up. (pointing at an object). There are two types of Endophors. An Anaphor is a scenario in which the referential entity/referent appears prior to its referencing pronoun. Anaphor Sentence Example: Ana bought a gown. She is in love with it. However, Cataphor is a condition in which the referent or entity occurs after its referencing pronoun. Exemple of Cataphor Sentence. “Ana bought the dress and didn’t realize it was torn.” “She” is the entity that appears before “Ana,” the referent entity. This is an example cataphor. A coreference cluster is also known as a chain of corefering words. Once we have a good understanding of the types of literature references, it’s important to understand its linguistic characteristics. These linguistic properties help us to understand coreference resolution in order to minimize error rates [1].. These properties can vary between languages depending on their rules. Before you attempt to resolve coreferences, make sure you have a good understanding of the grammar rules for that language. We will use English for clarity and simplicity in this article. Let me now explain each of these linguistic properties. Number Agreement simply means that the referencing expressions must agree in number. Gender Agreement means that the referencing words agree on gender. Let’s take a look at each. An example of Gender-Number Agreement: Analisa is a Google employee. It is her love for her job. Gender Agreement: “Analisa”, and “She”, agree on gender. They also agree on number. In other words, there is only one “Analisa”, and we use “She”, to refer to her, rather than using pronouns such as “he”, “they”, etc. It is a gender and number agreement. Number Agreement “Horses” is plural here. They agree to numbers. This agreement is called number. The Grammatical role is another property you should be aware of. The Grammatical role is a property that takes advantage of the inherent grammar of sentences and gives subject entities more importance than object entities. We assume an entity that is subject has a higher importance than an object entity. An example of Grammatical Role: “Ana works in an MNC together with Tia.” She works hard.” Grammatical Role We have “Ana”, “Tia”, and “She” as candidates for “She.” This sentence has “Ana”, the subject, and “Tia” the object. We consider “Ana”, keeping in mind its saliency to “She”, coreferent to “Tia.” Next, we need to think about Verb Semantics. When performing semantic analysis, some verbs are more likely to give meaning to one argument than others. An example of verb semantics “Ana saved Christa.” Ana condemned Christa. Ana was the architect of the project.” The use of the verb “helped,” in this sentence implies that Ana is more likely to be the architect than Christa. In the first sentence, Ana is referred to as “She”. The second sentence uses the verb “condemned”, which implies that Ana is more likely to be the architect of the project than Christa. In the second sentence, “She”, refers to “Christa”. Next, you need to be able to comprehend Selectional Restrictions. To determine the referent’s preference, this uses semantic knowledge of a sentence. Exemple of Selectional Restrictions “I ate the roasted bird in my pajamas, after it had been roasted for three hours in an oven.” Selectional restrictions Here are two potential referents to “it”, “chicken” and “pajamas”. Use of the verb
