Journal of Experimental and Theoretical Artificial Intelligence, 2011
This paper presents a series of experiments in automatically generating poetic texts. We confined... more This paper presents a series of experiments in automatically generating poetic texts. We confined our attention to the generation of texts which are syntactically well-formed, meet certain pre-specified patterns of metre, and broadly convey some given meaning. Such aspects can be formally defined, thus avoiding the complications of imagery and interpretation that are central to assessing more free forms of verse. Our implemented system, McGonagall, applies the genetic algorithm to construct such texts. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress. The conducted experiments broadly indicated that relatively meaningful text could be produced if the constraints on metre were relaxed, and precise metric text was possible with loose semantic constraints, but it was difficult to produce text which was both semantically coherent and of high quality metrically.
Rule Based Machine Translation (RBMT) and Statistical Machine translation (SMT) have different ap... more Rule Based Machine Translation (RBMT) and Statistical Machine translation (SMT) have different approach in performing translation task. RBMT uses linguistic rule between two languages which is built manually by human in general, whereas SMT uses co-occurrence statistic of word in parallel corpora. We combine those different approaches into Indonesian-English Hybrid Machine Translation (HMT) system to get the advantage from both kind of information. Initially, Indonesian text is inputted into RBMT. Then, the output will be edited by SMT to generate the final translation of English text. SMT is capable to do this because on the training process, it uses RBMT's output (English) as source material and real translation (English) as target material. Unavailability of ready to use Indonesian-English RBMT system becomes a challenge to do this research. Our study shows that SMT still outperforms HMT by 8.01% in average.
In this paper we describe a flexible approach to natural language generation that employs a stoch... more In this paper we describe a flexible approach to natural language generation that employs a stochastic hillclimbing search algorithm and an integrated architecture. We then discuss the benefits of this approach over existing, informative, goal-driven generation systems. We choose the generation of poetry as our research task domain, as it is a prime example of natural language that demands the degree of flexibility provided by our approach. Finally, we report and discuss results of our preliminary implementation work.
Poetry is a unique artifact of the human language faculty, with its defining feature being a stro... more Poetry is a unique artifact of the human language faculty, with its defining feature being a strong unity between content and form. Contrary to the opinion that the automatic generation of poetry is a relatively easy task, we argue that it is in fact an extremely difficult task that requires intelligence, world and linguistic knowledge, and creativity. We propose a model of poetry generation as a state space search problem, where a goal state is a text that satisfies the three properties of meaningfulness, grammaticality, and poeticness. We argue that ...
Abstract Although most research communities actively disseminate information about their projects... more Abstract Although most research communities actively disseminate information about their projects, publications, and members, there is no well-established standard for doing so, and thus the value of this data is reduced, eg in terms of accessibility, discoverability, and reusability. Linked Data is a method of publishing and linking structured data on the Web using RDF (Resource Description Framework). We propose LinkedLab: a Linked Data based solution for data management regarding research communities. The use of Linked ...
This paper describes an attempt to automatically map English words and concepts, derived from the... more This paper describes an attempt to automatically map English words and concepts, derived from the Princeton WordNet, to their Indonesian analogues appearing in an Indonesian lexicon. Using Latent Semantic Analysis (LSA), a semantic model is derived from an English-Indonesian parallel corpus. Given a particular word or concept, the semantic model is then used to identify its neighbours in a high-dimensional semantic space. Results from various experiments indicate that even with relatively small collections, LSA is able to discern some measure of implicit semantic knowledge from the training data.
This paper describes an experiment that attempts to automatically map English words and concepts,... more This paper describes an experiment that attempts to automatically map English words and concepts, derived from the Princeton WordNet, to their Indonesian analogues appearing in a widely-used Indonesian dictionary, using Latent Semantic Analysis (LSA). A bilingual semantic model is derived from an English-Indonesian parallel corpus. Given a particular word or concept, the semantic model is then used to identify its neighbours in a highdimensional semantic space. Results from various experiments indicate that for bilingual word mapping, LSA is consistently outperformed by the basic vector space model, i.e. where the full-rank word-document matrix is applied. We speculate that this is due to the fact that the 'smoothing' effect LSA has on the worddocument matrix, whilst very useful for revealing implicit semantic patterns, blurs the cooccurrence information that is necessary for establishing word translations.
Abstract In a question answering system (QAS), question analysis component has an important task ... more Abstract In a question answering system (QAS), question analysis component has an important task to determine the expected answer type (EAT) of a given question. Many QAS's rely their question analysis performance on manually developed patterns, such as in Open Ephyra (OE), one of a state of the art freely available QAS. Recently, there are a number of studies which investigated the influence of statistical relational framework to learn question-answer pairs in particular component of a QAS. In this study, we propose an ...
Journal of Experimental and Theoretical Artificial Intelligence, 2012
This paper presents a series of experiments in automatically generating poetic texts. We confined... more This paper presents a series of experiments in automatically generating poetic texts. We confined our attention to the generation of texts which are syntactically well-formed, meet certain pre-specified patterns of metre, and broadly convey some given meaning. Such aspects can be formally defined, thus avoiding the complications of imagery and interpretation that are central to assessing more free forms of verse. Our implemented system, McGonagall, applies the genetic algorithm to construct such texts. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress. The conducted experiments broadly indicated that relatively meaningful text could be produced if the constraints on metre were relaxed, and precise metric text was possible with loose semantic constraints, but it was difficult to produce text which was both semantically coherent and of high quality metrically.
Children who are deprived of language play are likely to experience impaired linguistic developme... more Children who are deprived of language play are likely to experience impaired linguistic development and this can in turn inhibit communicative and social skills (Donahue & Bryan 84). A software tool which will permit a child with language and/or speech impairments to explore language in an enjoyable way, by manipulating words and phrases into simple jokes, is being built to address this issue. The automated joke-construction is based on the ideas used in the JAPE program, which could form simple punning riddles broadly similar to those in published joke books (Binsted et al. 97). JAPE was a rather basic prototype, with no real user interface, almost no facilities for guiding the joke-construction, and a response time (hours) wholly unsuitable for an interactive system. Following the philosophy of user-centred design (Vredenburg et al. 02), an extensive consultation was undertaken with domain experts (speech and language therapists) and potential users to ascertain what a joke-building system for children with disabilities should provide (O'Mara et al. 04, Manurung et al. 05).
In this paper we describe the difficulties of poetry generation, particularly in contrast to trad... more In this paper we describe the difficulties of poetry generation, particularly in contrast to traditional informative natural language generation. We then point out deficiencies of previous attempts at poetry generation, and propose a stochastic hillclimbing search model which addresses these deficiencies. We present both conceptual and implemented details of the most important aspects of such a model, the evaluation and evolution functions. Finally, we report and discuss results of our preliminary implementation work.
This paper explores the feasibility of implementing an evolutionary algorithm for text structurin... more This paper explores the feasibility of implementing an evolutionary algorithm for text structuring using the heuristic of continuity as a fitness function, chosen over other more complicated metrics of text coherence. Using MCGONAGALL as our experimental platform, we show that by employing an elitist strategy for stochastic search it is possible to quickly reach the global optimum of minimal violations of continuity.
A WordNet is a useful lexical resource where specific senses of words are clustered together into... more A WordNet is a useful lexical resource where specific senses of words are clustered together into synonym sets, and semantic relationships between these sets are specified. This paper describes an ongoing project to create an Indonesian WordNet using the expand model approach, i.e. by mapping existing WordNet entries to Indonesian word sense definitions. We discuss some issues encountered during the development of a web-based application that facilitates this mapping. Some initial results of an experiment to measure the inter-annotator reliability are presented, along with a brief discussion on the constructed Indonesian WordNet.
The Faculty of Computer Science, University of Indonesia, has been conducting research concerning... more The Faculty of Computer Science, University of Indonesia, has been conducting research concerning the computational modelling and processing of various aspects of bahasa Indonesia since the mid-1990s. This paper serves as a historical account of that work, as a convenient bibliographic reference, and as a platform from which to identify future research.
Sentiment analysis is the automatic classification of the overall opinion conveyed by a text towa... more Sentiment analysis is the automatic classification of the overall opinion conveyed by a text towards its subject matter. This paper discusses an experiment in the sentiment analysis of of a collection of movie reviews that have been automatically translated to Indonesian. Following [1], we employ three well known classification techniques: naive bayes, maximum entropy, and support vector machines, employing unigram presence and frequency values as the features. The translation is achieved through machine translation and simple word substitutions based on a bilingual dictionary constructed from various online resources. Analysis of the Indonesian translations yielded an accuracy of up to 78.82%, still short of the accuracy for the English documents (80.09%), but satisfactorily high given the simple translation approach.
This paper presents a model of deep syntactic and semantic processing to support question-answeri... more This paper presents a model of deep syntactic and semantic processing to support question-answering for Bahasa Indonesia. Starting from an existing unification-based grammar, we specify lexical semantics for each lexeme and semantic attachment rules for each grammar rule using lambdacalculus notation. This approach enables us to obtain semantic representations of Indonesian sentences in the form of first order logic literals. These representations are used by a question-answering module which stores declarative sentences as facts in a knowledge-base, and answers queries by unifying the question representation with known facts. We discuss an implemented prototype of this model using Prolog, and conclude with the remaining work that has to be done to realise our goal of a fully-fledged question-answering system for Bahasa Indonesia.
Abstract In this paper, Aitana, a Developmental Cognitive Artifact developed by Marques (2006), i... more Abstract In this paper, Aitana, a Developmental Cognitive Artifact developed by Marques (2006), is used to produce “conceptual” representations of the best known Cellular Automata (CA) rules that perform the Density Classification Task (DCT)(see eg Packard, 1984, 1988; Mitchell et al., 1993). The resulting redescriptions are then analysed from the viewpoint of their similarity, in terms of their cell state-update functions. We show that these new conceptual representations can reveal knowledge about these rules that is not accessible ...
As part of a project to construct an interactive program which will encourage children to play wi... more As part of a project to construct an interactive program which will encourage children to play with language by building jokes, we have developed a large lexical database, closely based on WordNet. As well as the standard WordNet information about part of speech, synonymy, hyponymy, etc, we have added phonetic representations and symbolic links allowing attachment of pictures. All information is represented in a relational database, allowing powerful searches using SQL via a Java API. The lexicon has a facility to label subsets of the lexicon with symbolic names, and we are working to incorporate some educationally relevant word lists as sublexicons. This should also allow us to improve the familiarity ratings which the lexicon assigns to words.
Journal of Experimental and Theoretical Artificial Intelligence, 2011
This paper presents a series of experiments in automatically generating poetic texts. We confined... more This paper presents a series of experiments in automatically generating poetic texts. We confined our attention to the generation of texts which are syntactically well-formed, meet certain pre-specified patterns of metre, and broadly convey some given meaning. Such aspects can be formally defined, thus avoiding the complications of imagery and interpretation that are central to assessing more free forms of verse. Our implemented system, McGonagall, applies the genetic algorithm to construct such texts. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress. The conducted experiments broadly indicated that relatively meaningful text could be produced if the constraints on metre were relaxed, and precise metric text was possible with loose semantic constraints, but it was difficult to produce text which was both semantically coherent and of high quality metrically.
Rule Based Machine Translation (RBMT) and Statistical Machine translation (SMT) have different ap... more Rule Based Machine Translation (RBMT) and Statistical Machine translation (SMT) have different approach in performing translation task. RBMT uses linguistic rule between two languages which is built manually by human in general, whereas SMT uses co-occurrence statistic of word in parallel corpora. We combine those different approaches into Indonesian-English Hybrid Machine Translation (HMT) system to get the advantage from both kind of information. Initially, Indonesian text is inputted into RBMT. Then, the output will be edited by SMT to generate the final translation of English text. SMT is capable to do this because on the training process, it uses RBMT's output (English) as source material and real translation (English) as target material. Unavailability of ready to use Indonesian-English RBMT system becomes a challenge to do this research. Our study shows that SMT still outperforms HMT by 8.01% in average.
In this paper we describe a flexible approach to natural language generation that employs a stoch... more In this paper we describe a flexible approach to natural language generation that employs a stochastic hillclimbing search algorithm and an integrated architecture. We then discuss the benefits of this approach over existing, informative, goal-driven generation systems. We choose the generation of poetry as our research task domain, as it is a prime example of natural language that demands the degree of flexibility provided by our approach. Finally, we report and discuss results of our preliminary implementation work.
Poetry is a unique artifact of the human language faculty, with its defining feature being a stro... more Poetry is a unique artifact of the human language faculty, with its defining feature being a strong unity between content and form. Contrary to the opinion that the automatic generation of poetry is a relatively easy task, we argue that it is in fact an extremely difficult task that requires intelligence, world and linguistic knowledge, and creativity. We propose a model of poetry generation as a state space search problem, where a goal state is a text that satisfies the three properties of meaningfulness, grammaticality, and poeticness. We argue that ...
Abstract Although most research communities actively disseminate information about their projects... more Abstract Although most research communities actively disseminate information about their projects, publications, and members, there is no well-established standard for doing so, and thus the value of this data is reduced, eg in terms of accessibility, discoverability, and reusability. Linked Data is a method of publishing and linking structured data on the Web using RDF (Resource Description Framework). We propose LinkedLab: a Linked Data based solution for data management regarding research communities. The use of Linked ...
This paper describes an attempt to automatically map English words and concepts, derived from the... more This paper describes an attempt to automatically map English words and concepts, derived from the Princeton WordNet, to their Indonesian analogues appearing in an Indonesian lexicon. Using Latent Semantic Analysis (LSA), a semantic model is derived from an English-Indonesian parallel corpus. Given a particular word or concept, the semantic model is then used to identify its neighbours in a high-dimensional semantic space. Results from various experiments indicate that even with relatively small collections, LSA is able to discern some measure of implicit semantic knowledge from the training data.
This paper describes an experiment that attempts to automatically map English words and concepts,... more This paper describes an experiment that attempts to automatically map English words and concepts, derived from the Princeton WordNet, to their Indonesian analogues appearing in a widely-used Indonesian dictionary, using Latent Semantic Analysis (LSA). A bilingual semantic model is derived from an English-Indonesian parallel corpus. Given a particular word or concept, the semantic model is then used to identify its neighbours in a highdimensional semantic space. Results from various experiments indicate that for bilingual word mapping, LSA is consistently outperformed by the basic vector space model, i.e. where the full-rank word-document matrix is applied. We speculate that this is due to the fact that the 'smoothing' effect LSA has on the worddocument matrix, whilst very useful for revealing implicit semantic patterns, blurs the cooccurrence information that is necessary for establishing word translations.
Abstract In a question answering system (QAS), question analysis component has an important task ... more Abstract In a question answering system (QAS), question analysis component has an important task to determine the expected answer type (EAT) of a given question. Many QAS's rely their question analysis performance on manually developed patterns, such as in Open Ephyra (OE), one of a state of the art freely available QAS. Recently, there are a number of studies which investigated the influence of statistical relational framework to learn question-answer pairs in particular component of a QAS. In this study, we propose an ...
Journal of Experimental and Theoretical Artificial Intelligence, 2012
This paper presents a series of experiments in automatically generating poetic texts. We confined... more This paper presents a series of experiments in automatically generating poetic texts. We confined our attention to the generation of texts which are syntactically well-formed, meet certain pre-specified patterns of metre, and broadly convey some given meaning. Such aspects can be formally defined, thus avoiding the complications of imagery and interpretation that are central to assessing more free forms of verse. Our implemented system, McGonagall, applies the genetic algorithm to construct such texts. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress. The conducted experiments broadly indicated that relatively meaningful text could be produced if the constraints on metre were relaxed, and precise metric text was possible with loose semantic constraints, but it was difficult to produce text which was both semantically coherent and of high quality metrically.
Children who are deprived of language play are likely to experience impaired linguistic developme... more Children who are deprived of language play are likely to experience impaired linguistic development and this can in turn inhibit communicative and social skills (Donahue & Bryan 84). A software tool which will permit a child with language and/or speech impairments to explore language in an enjoyable way, by manipulating words and phrases into simple jokes, is being built to address this issue. The automated joke-construction is based on the ideas used in the JAPE program, which could form simple punning riddles broadly similar to those in published joke books (Binsted et al. 97). JAPE was a rather basic prototype, with no real user interface, almost no facilities for guiding the joke-construction, and a response time (hours) wholly unsuitable for an interactive system. Following the philosophy of user-centred design (Vredenburg et al. 02), an extensive consultation was undertaken with domain experts (speech and language therapists) and potential users to ascertain what a joke-building system for children with disabilities should provide (O'Mara et al. 04, Manurung et al. 05).
In this paper we describe the difficulties of poetry generation, particularly in contrast to trad... more In this paper we describe the difficulties of poetry generation, particularly in contrast to traditional informative natural language generation. We then point out deficiencies of previous attempts at poetry generation, and propose a stochastic hillclimbing search model which addresses these deficiencies. We present both conceptual and implemented details of the most important aspects of such a model, the evaluation and evolution functions. Finally, we report and discuss results of our preliminary implementation work.
This paper explores the feasibility of implementing an evolutionary algorithm for text structurin... more This paper explores the feasibility of implementing an evolutionary algorithm for text structuring using the heuristic of continuity as a fitness function, chosen over other more complicated metrics of text coherence. Using MCGONAGALL as our experimental platform, we show that by employing an elitist strategy for stochastic search it is possible to quickly reach the global optimum of minimal violations of continuity.
A WordNet is a useful lexical resource where specific senses of words are clustered together into... more A WordNet is a useful lexical resource where specific senses of words are clustered together into synonym sets, and semantic relationships between these sets are specified. This paper describes an ongoing project to create an Indonesian WordNet using the expand model approach, i.e. by mapping existing WordNet entries to Indonesian word sense definitions. We discuss some issues encountered during the development of a web-based application that facilitates this mapping. Some initial results of an experiment to measure the inter-annotator reliability are presented, along with a brief discussion on the constructed Indonesian WordNet.
The Faculty of Computer Science, University of Indonesia, has been conducting research concerning... more The Faculty of Computer Science, University of Indonesia, has been conducting research concerning the computational modelling and processing of various aspects of bahasa Indonesia since the mid-1990s. This paper serves as a historical account of that work, as a convenient bibliographic reference, and as a platform from which to identify future research.
Sentiment analysis is the automatic classification of the overall opinion conveyed by a text towa... more Sentiment analysis is the automatic classification of the overall opinion conveyed by a text towards its subject matter. This paper discusses an experiment in the sentiment analysis of of a collection of movie reviews that have been automatically translated to Indonesian. Following [1], we employ three well known classification techniques: naive bayes, maximum entropy, and support vector machines, employing unigram presence and frequency values as the features. The translation is achieved through machine translation and simple word substitutions based on a bilingual dictionary constructed from various online resources. Analysis of the Indonesian translations yielded an accuracy of up to 78.82%, still short of the accuracy for the English documents (80.09%), but satisfactorily high given the simple translation approach.
This paper presents a model of deep syntactic and semantic processing to support question-answeri... more This paper presents a model of deep syntactic and semantic processing to support question-answering for Bahasa Indonesia. Starting from an existing unification-based grammar, we specify lexical semantics for each lexeme and semantic attachment rules for each grammar rule using lambdacalculus notation. This approach enables us to obtain semantic representations of Indonesian sentences in the form of first order logic literals. These representations are used by a question-answering module which stores declarative sentences as facts in a knowledge-base, and answers queries by unifying the question representation with known facts. We discuss an implemented prototype of this model using Prolog, and conclude with the remaining work that has to be done to realise our goal of a fully-fledged question-answering system for Bahasa Indonesia.
Abstract In this paper, Aitana, a Developmental Cognitive Artifact developed by Marques (2006), i... more Abstract In this paper, Aitana, a Developmental Cognitive Artifact developed by Marques (2006), is used to produce “conceptual” representations of the best known Cellular Automata (CA) rules that perform the Density Classification Task (DCT)(see eg Packard, 1984, 1988; Mitchell et al., 1993). The resulting redescriptions are then analysed from the viewpoint of their similarity, in terms of their cell state-update functions. We show that these new conceptual representations can reveal knowledge about these rules that is not accessible ...
As part of a project to construct an interactive program which will encourage children to play wi... more As part of a project to construct an interactive program which will encourage children to play with language by building jokes, we have developed a large lexical database, closely based on WordNet. As well as the standard WordNet information about part of speech, synonymy, hyponymy, etc, we have added phonetic representations and symbolic links allowing attachment of pictures. All information is represented in a relational database, allowing powerful searches using SQL via a Java API. The lexicon has a facility to label subsets of the lexicon with symbolic names, and we are working to incorporate some educationally relevant word lists as sublexicons. This should also allow us to improve the familiarity ratings which the lexicon assigns to words.
Uploads
Papers by Ruli Manurung