This thesis presents a knowledge-poor pronoun resolution system for Turkish which resolves third person personal pronouns and possessive pronouns. The system is knowledge-poor in the sense that it makes use of limited linguistic and semantic knowledge to resolve the pronouns. As pronoun resolution proposals for languages like English, French and Spanish, the core of the system is the constraints and preferences which are determined empirically. The system has four modules: sentence splitting, pronoun extraction, forming the list of candidate antecedents and determination of the antecedent. It takes a Turkish text as input and rewrites this text with the considered pronouns replaced with their proposed antecedents. In order to compare the success rate of the system, two different baseline algorithms are implemented. The original system is tested against these baseline algorithms on two sample Turkish texts from different sources. Some suggestions to improve the success rate of the system and to extend the domain of the system are also presented.

This thesis develops an approach to extract the social relation between characters in literary text to create a social network. The approach uses co-occurrences of named entities, keywords associated with the named entities, and the dependency relations that exist between the named entities to construct the network. Literary texts contain a large amount of pronouns to represent the named entities, to resolve the antecedents of pronouns, a pronoun resolution system is implemented based on a standard pronoun resolution algorithm. The results indicate that the pronoun resolution system finds the correct named entity in 60,4\% of all cases. The social network is evaluated by comparing character importance rankings based on graph properties with an independently human generated importance rankings. The generated social networks correlate moderately to strongly with the independent character ranking.

In this thesis, I explored several neural network-based models to resolve the issues of zero pronoun in Chinese English translation tasks. I reviewed previous work that attempts to take the resolution as a classification task, such as determining if a candidate in a given set is the antecedent of a zero pronoun, which can be categorized as rule-based and supervised methods. Existing methods either did not take the relationship between potential zero pronoun candidates into consideration or did not fully utilize attention to zero pronoun representations. In my experiments, I investigated attention-based neural network models as well as its application in reinforcement learning environment building on an existing neural model. In particular, I integrated an LSTM-tree-based module into the attention network, which encodes syntax information for zero pronoun resolution tasks. In addition, I apply Bi-Attention layers between modules to interactively learn the syntax and semantic alignment. Furthermore, I leveraged a reinforcement learning framework to fine-tune the proposed model, and experiment with different encoding strategies, i.e., FastText, BERT, and trained RNN-based embedding. I found that attention-based model with LSTM-tree- based module, fine-tuned under reinforcement learning framework that utilized FastText embedding achieves the best performance, superior to the baseline models. I evaluated the model performance on different categories of resources, of which FastText shows great potential in encoding web blog text.

This report describes a quantitative analysis performed to compare two different methods on the task of pronoun resolution for Swedish. The first method, an implementation of Mitkov’s algorithm, is a heuristic-based method — meaning that the resolution is determined by a number of manually engineered rules regarding both syntactic and semantic information. The second method is data-driven — a Support Vector Machine (SVM) using dependency trees and word embeddings as features. Both methods are evaluated on an annotated corpus of Swedish news articles which was created as a part of this thesis. SVM-based methods significantly outperformed the implementation of Mitkov’s algorithm. The best performing SVM model relies on tree kernels applied to dependency trees. The model achieved an F1-score of 0.76 for the positive class and 0.9 for the negative class, where positives are pairs of pronoun and noun phrase that corefer, and negatives are pairs that do not corefer. / Rapporten beskriver en kvantitativ analys som genomförts för att jämföra två olika metoder för automatisk pronomenbestämning på svenska. Den första metoden, en implementation av Mitkovs algoritm, är en heuristisk metod vilket innebär att pronomenbestämningen görs med ett antal manuellt utformade regler som avser att fånga både syntaktisk och semantisk information. Den andra metoden är datadriven, en stödvektormaskin (SVM) som använder dependensträd och ordvektorer som särdrag. Båda metoderna utvärderades med hjälp av en annoterad datamängd bestående av svenska nyhetsartiklar som skapats som en del av denna avhandling. Den datadrivna metoden överträffade Mitkovs algoritm. Den SVM-modell som ger bäst resultat bygger på trädkärnor som tillämpas på dependensträd. Modellen uppnådde ett F1-värde på 0.76 för den positiva klassen och 0.9 för den negativa klassen, där de positiva datapunkterna utgörs av ett par av pronomen och nominalfras som korefererar, och de negativa datapunkterna utgörs av par som inte korefererar.