Includes bibliographical references.

Automated Planning of Children’s Stories using Causal Links, Agents and Commonsense Ontology -- Community Structure Detection and Analysis in Disaster Related Tweets -- Composer Classification of Filipino Song Lyrics Using Machine Learning -- Disaster-Related Participant Tweet Identification Using SVM -- Generating Text Descriptions in the Alex Interactive Storytelling System using a Semantic Ontology -- Some Observations on the Sociophonetics of Kapampangan -- Toward an Optimal Multilingual Natural Language Generator: Deep Source Analysis and Shallow Target Analysis.

[Article Title: Automated Planning of Children’s Stories using Causal Links, Agents and Commonsense Ontology/ Ethel Chua Joy Ong and Jasmine Irene Ng, p.1-10]

Abstract: Building computer systems that have the ability to generate human understandable stories has been the subject of research in the field of computational narrative, as these can find applications in education and entertainment. But story planning, which involves the identification of a story plot structure and the production of a coherent and interesting sequence of events that lead story characters to perform actions and experience emotions in order to achieve their goals, remain a challenging task.

In this paper, we present a comparative discussion and analysis of the approaches employed by Picture Books 2 (PB2) and its variants, PB2 Planning Agents (PB2-PA) and PB2 Concept Net (PB2-CN), to produce children's stories of the fable form. The comparison is aimed at determining whether the enhancement of the storytelling knowledge and the use of agents that can reason over this knowledge during story planning would lead to the generation of stories that contain more cohesive story events.

Specifically, two forms of enhancement were made; first, by using pre- and post-conditions as additional criteria in the selection of candidate events; and second, by using existing language resources to supply the necessary commonsense knowledge needed by the planner.
Comparative evaluation of the three systems using the criteria on coherence and cohesion, story elements and content showed that PB2-PA received the highest overall average score of 4.25 out of 5. This is followed by PB2 which garnered an average score of 3.81. PB2-CN garnered the lowest average score of 3.29. The use of three agents (character, plot and world agents) in PB2-PA led to the generation of better story plans containing character actions that are more consistent and directed to the selected theme. The use of existing resources in PB2-CN, though appropriate and relevant to the identified story themes, is found to be insufficient to support the storytelling knowledge requirements of the planner.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18

[Article Title: Community Structure Detection and Analysis in Disaster Related Tweets/ Harriet Angelie A. Gonzales and Kurt Junshean P. Espinosa, p. 11-16]

Abstract : One of the considered principal disasters that hit the Philippines almost year round is flooding. At the occurrence of such floods, social media – Twitter for instance – serve as communication outlet between users rendering them significant in information gathering and dissemination. This study aims to determine the significance of social networks when it comes to disaster information by analyzing community structures formed from different graph relationships and comparing it to actual patterns of flood affected areas of the same timeframe. This paper analyzes the properties of the community structure detected among nodes in a social network graph formed among Filipino Twitter users who tweeted about flood. Interaction relationship graph was created wherein an edge is formed between two users if user A mentions user B. Seventy-seven communities with more than ten nodes were detected. However, nodes belonging in the same community did not show similarities with each other.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp11-16%202014

[Article Title: Composer Classification of Filipino Song Lyrics Using Machine Learning/ Oliver Isaac L. Chan, Patrick Joseph M. Sadornas, Rafael A. Cabredo, and Charibeth K. Cheng, p. 17-23]

Abstract: We implemented a machine learning technique using artificial neural network to perform composer classification (authorship attribution) of songs with lyrics primarily written in Filipino. We used features based on function words, character n-grams, and song-specific features extracted from 98 song lyrics written by three composers and used a multilayer perceptron to model the learning algorithm for automatically classifying song lyrics according to its likely composer. Compared to classification of longer literary materials such as novels, author attribution of short texts such as Filipino song lyrics is generally a more difficult machine learning task. By combining the function words and character n-grams, we can achieve an average classification accuracy of 81.02%. In contrast, a random guess on a song’s composer would have an average of 33.33% chance of correct classification. Our results demonstrate a successful application of author attribution methods to short Filipino texts.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp17-23%202014

[Article Title: Disaster-Related Participant Tweet Identification Using SVM/ Lemuel John Beduya and Kurt Junshean Espinosa, p. 24-33]

Abstract: The Philippines is a hot bed of disasters: earthquakes, flooding, and fires often occur in the countryMoreover, in this part of the globe, majority of the populace are very attuned to social media with almost everyone who are either Twitter or Facebook users. This study took advantage of that and used Twitter in identifying the disaster-related participant tweets by Filipino users in the Philippines. This study will aid the Philippine government and other concerned organizations in their disaster management plans. In view of this, a multi-level binary classification on tweets was implemented using SVM. Specifically, the first level identified a tweet if it is a disaster-related participant tweet or not. The next level identified the type of disaster the participant is experiencing which can be flood, earthquake or others (fire, landslide, etc). In order to yield the best model for each data set, a 10-fold cross-validation was performed. The process yielded a model for each data set with an F1score of 0.73, 0.83 and 0.72 for disaster-related, flood-related and earthquake-related participant tweets respectively. The results of the study showed that it is indeed possible to identify participant tweets of any type of disaster in Twitter using SVM. Furthermore, this study can be used as a starting point in examining if it is possible to identify the disaster-prone areas in the Philippines using Twitter.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp24-33%202014

[Article Title: Generating Text Descriptions in the Alex Interactive Storytelling System using a Semantic Ontology/ Rancelli Jane Roxas, Dolleen Lour Huang, Bruce Elmer Peralta, Sashmir Yap, and Ethel Chua Joy Ong, p. 34-43]

Abstract: Semantic ontologies of commonsense concepts have been used to provide the necessary knowledge needed by computer systems to perform various tasks, such as in automatic story generation, interactive scavenger hunt games, virtual museum and storytelling. This paper presents our work in the development of a semantic ontology to support the generation of narrative text in the Alex interactive storytelling system. The text generated involves simple textual description of the objects found in the virtual world, the current options that are presented to the user to move the story forward, and a simple text feedback about the choices made by the user.

Results from user evaluation showed that the semantic ontology can be used to model concepts representing the various elements of a story. These include characters, locations, objects and their attributes, as well as story events. However, the text generation process can still be improved specifically in the production of text that describes the importance of acquiring an item, keeping track of the progress of the user towards fulfilling the goal of the story, and detailing how the previous and current choices of the user affected the story flow.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp34-43%202014

[Article Title: Some Observations on the Sociophonetics of Kapampangan/ Anne Grace H. Peralta and Layton Jan M. Tee, p. 44-54]

Abstract: Kapampangan, one of Philippines’ most spoken languages and like all languages, has its own distinct characteristics that make it unique from other languages. This paper endeavors to investigate whether the literature written, primarily by Del Corro [6, 7]and Gonzalez [10]about the phonetics of the language are accurate. These phonetic observations can contribute to how people can view or perceive the language sociophonetically. These sociophonetic features that were observed and provided include the vowel inventory of Kapampangan, the loss of the /h/ sound in Kapampangan variants, vowel lowering in final position, the true description of Del Corro’s circumflexed a, glides, monophthongization and the occurrence of glottal stop. A preliminary study on Kapampangan intonation, intonation units and prosodic patterns were also discussed using the Tobi method. These were the identied features from past written literature that may be a basis to sociophonetically distinguish the language from other Philippine-type languages. With the use of modern acoustic technology, the Praat software, this paper will attempt to give light on this observations and will try to verify whether these observations are accurate or not and in the end, aims to contribute to the growing literature of the language.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp44-54%202014

[Article Title: Toward an Optimal Multilingual Natural Language Generator: Deep Source Analysis and Shallow Target Analysis/ Tod Allman, Stephen Beale, and Richard Denton, p. 55-63]

Abstract: Linguist’s Assistant (LA) is a large scale multilingual natural language generator (NLG) designed and developed entirely from a linguist’s perspective. The system incorporates extensive typological, semantic, syntactic, and discourse research into its semantic representational system and its transfer and synthesizing grammars. LA has been tested with English, Korean, Kewa (Papua New Guinea), and Jula (Cote d’Ivoire), and proof of concept lexicons and grammars have been developed for a variety of other languages. The system has generated initial draft translations of texts in each of the test languages, and when experienced mother-tongue translators edit those drafts into publishable texts, their productivity is typically quadrupled when compared with manual translation.

An optimal NLG will be able to generate high quality texts in a wide variety of languages with minimal knowledge of the target language grammars. In order to increase the quality of the drafts generated by LA, deep source analysis techniques have been adopted. And in order to minimize the target language knowledge that is required to generate the drafts, a new approach to grammar development has been designed into LA’s synthesizing grammar. This paper will:1) summarize the major components of the generation system, 2) describe several of the source analysis techniques that have been adopted during the development of LA’s semantic representations, and 3) present an example of the new type of synthesizing rule that was added to LA’s grammar. The adoption of deep source analysis techniques combined with shallow target analysis has proven to be a very efficient model.

https://pcj.csp.org.ph/index.php/pcj/issue/view/18/PCJ%20V9%20N1%20pp55-63%202014