Nobuo Inui

Some thoughts on hill-climbing method

A hill-climbing method is popular for optimization problems. A problem of a general hillclimbing method is that the process of getting solutions is dependent on the nature of a problem and the way of determining initial states and the next states, etc. For this, a person who considers a hillclimbing method must carefblly observe problems, because there is no guarantee of finding optimal solution for a limited period. Especially in industrial use:.. ticvelopers hesitate about applying hillclimbin; nlethods to their application for this reason. In adotbun, it is difficult to apply some methods like the steepest decent method for searching the next state effectively, as objective functions become complicated. In this paper, we describe a solution against time-constraints and difficulties of finding the next states in a hill-climbing method by a general transform of objective functions. In addition. some thoughts for improving a hill-climbing method are described.

The Design and Experiment of an Evaluation Function for User Interaction Cost in the Interactive Semantic Disambiguation

Abstract In this paper, we will design a function which evaluates the cost of user interaction that resolves the semantic ambiguity in natural language processing system. Through user interaction, the system shows some example sentences that express each meaning to let the user select one from them. We define the evaluation function based on semantic and syntactic features of verbs.

Generating a set of rules to determine honorific expression using decision tree learning

In Japanese language, the speaker must choose suitable honorific expressions depending on many factors. The computer system should imitate this mechanism to make a natural Japanese sentence. We made a system to determine a suitable expression and named it honorific expression determining system (HEDS). It generates a set of rules to determine suitable honorific expression automatically, by decision tree learning. The system HEDS determines one out of the three classes for an input sentence: the respect expression, the modesty expression and the non-honorific expression and determines what expression the verb is. We calculated the accuracy of HEDS using the cross validation method and it was up to 74.88%.

Minimizing state transition model for multiclassification by mixed-integer programming

This paper proposes a state transition (ST) model as a classifier and its generalization by the minimization. Different from previous works using statistical methods, tree-based classifiers and neural networks, we use a ST model which determines classes of strings. Though an initial ST model only accepts given strings, the minimum ST model can accepts various strings by the generalization. We use a minimization algorithm by Mixed-Integer Linear Programming (MILP) approach. The MILP approach guarantees a minimum solution. Experiment was done for the classification of pseudo-strings. Experimental results showed that the reduction ratio from an initial ST model to the minimal ST model becomes small, as the number of examples increases. However, a current MILP solver was not feasible for large scale ST models in our formalization.

Combination of an Automatic and an Interactive Disambiguation Method

In natural language processing, many methods have been proposed to solve the ambiguity problems. In this paper, we propose a technique to combine a method of interactive disambiguation and automatic one for ambiguous words. The characteristic of our method is that the accuracy of the interactive disambiguation is considered. The method solves the two following problems when combining those disambiguation methods: (1) when should the interactive disambiguation be executed? (2) which ambiguous word should be disambiguated when more than one ambiguous words exist in a sentence? Our method defines the condition of executing the interaction with users and the order of disambiguation based on the strategy where the accuracy of the result is maximized, considering the accuracy of the interactive disambiguation and automatic one. Using this method, user interaction can be controlled while holding the accuracy of results.