Kohichi Sayama

Reducing excessive amounts of data: multiple web queries for generation of pun candidates

Humor processing is still a less studied issue, both in NLP and AI. In this paper we contribute to this field. In our previous research we showed that adding a simple pun generator to a chatterbot can significantly improve its performance. The pun generator we used generated only puns based on words (not phrases). In this paper we introduce the next stage of the systems development-- an algorithm allowing generation of phrasal pun candidates. We show that by using only the Internet (without any handmade humor-oriented lexicons), it is possible to generate puns based on complex phrases. As the output list is often excessively long, we also propose a method for reducing the number of candidates by comparing two web-query-based rankings. The evaluation experiment showed that the system achieved an accuracy of 72.5% for finding proper candidates in general, and the reduction method allowed us to significantly shorten the candidates list. The parameters of the reduction algorithm are variable, so that the balance between the number of candidates and the quality of output can be manipulated according to needs.

Simile or Not Simile

In this paper we propose a method of automatic distinction between two types of formally identical expressions in Japanese: similes and “metonymical comparisosn”, i.e. literal comparisons that include metonymic relations between elements. Expression like “kujira no you na chiisai me” can be translated into English as “eyes small as whale’s”, while in Japanese, due to the lack of possessive case, it can be misunderstood as “eyes small as a whale”. The reason behind this is the presence of metonymic relation between components of such expressions. In the abovegiven example the word “whale” is a metonymy and represents “whale’s eye”. This is naturally understandable for humans, although formally difficult to detect by automatic algorithms, as both types of expressions (similes and metonymical comparisons) realize the same template. In this work we present a system able to distinguish between these two types of expressions. The system takes a Japanese expression as input and uses the Internet to check possessive relations between its elements. We propose a method of calculating a score based on co-occurrence of source and target pairs in Google (e.g. “whale’s eye”). Evaluation experiment showed that the system distinguishes between similes and metonimical comparisons with the accuracy of 74 %. We discuss the results and give some ideas for the future.

NLP Oriented Japanese Pun Classification

In this paper we describe a phonetic classification of Japanese puns (dajare). Basing on real life examples gathered from available sources (books, Internet), we divided Japanese puns into 12 groups with numerous subgroups, according to phonetic changes that occur within them. This classification was prepared for the NLP purpose, i.e. to be used in humor processing. Its usefulness was shown in a research project, aimed at constructing a humor-equipped conversational system for Japanese.

Data filtering in humor generation: comparative analysis of hit rate and co-occurrence rankings as a method to choose usable pun candidates

In this paper we propose a method of filtering excessive amount of textual data acquired from the Internet. In our research on pun generation in Japanese we experienced problems with extensively long data processing time, caused by the amount of phonetic candidates generated (i.e. phrases that can be used to generate actual puns) by our system. Simple, naive approach in which we take into considerations only phrases with the highest occurrence in the Internet, can effect in deletion of those candidates that are actually usable. Thus, we propose a data filtering method in which we compare two Internet-based rankings: a co-occurrence ranking and a hit rate ranking, and select only candidates which occupy the same or similar positions in these rankings. In this work we analyze the effects of such data reduction, considering 1 cases: when the candidates are on exactly the same positions in both rankings, and when their positions differ by 1, 2, 3 and 4. The analysis is conducted on data acquired by comparing pun candidates generated by the system (and filtered with our method) with phrases that were actually used in puns created by humans. The results show that the proposed method can be used to filter excessive amounts of textual data acquired from the Internet.