A Computational Approach to Re-Interpretation: Generation of Emphatic Poems Inspired by Internet Blogs

Transcription

1 Modeling Changing Perspectives Reconceptualizing Sensorimotor Experiences: Papers from the 2014 AAAI Fall Symposium A Computational Approach to Re-Interpretation: Generation of Emphatic Poems Inspired by Internet Blogs Joanna Misztal and Bipin Indurkhya Faculty of Mathematics and Computer Science, Jagiellonian University, Krakow, Poland Computer Science Department AGH University of Science and Technology, Krakow, Poland Abstract We present a system that produces emotionally rich poetry inspired by personalized and empathic interpretation of text, particularly Internet blogs. Our implemented system is based on the blackboard architecture, and generates poetry from a theme that it considers the most inspiring. It also incorporates a model of emotions with an individual optimism rate that defines an affective state. The poems produced by the system contain emotional expressions that describe these feelings. We explain how the system re-conceptualizes the text by the empathic interpretation of its content. We also present how the blackboard architecture may support divergent problem solving in the field of computational creativity. We describe the system architecture and the generation algorithm followed by some illustrative results. Finally, we mention possible continuation of this work by incorporating other language generating systems as well as human experts in the blackboard architecture. Introduction The act of creating art may be considered as a reconceptualization of an input stimuli (inspiration) into an artistic artifact. A painter, when she or he creates a portrait, does not necessarily replicate a precise image of the painted object. The created work is the reflection of the author s impressions and feelings inspired by some real phenomena. Other artists work in a similar manner. They get some inspiration from the environment, interpret this information in their own way, often dependent on their emotional state, and create artifacts that express their point of view and feelings evoked by the inspiring object or situation. Our goal is to model this creative process of reconceptualization during the artists work. We describe here a system that produces poems inspired by the input text. For the experiment, we use fragments of Internet blog posts, which usually carry some emotional content. Our system analyzes the text first, trying to find the most interesting term. It rates phrases according to the number of diverse associations it can derive from each expression. Then we use the computational empathy (Misztal and Indurkhya 2014b) to define the emotional state. This state is calculated using an Copyright c 2014, Association for the Advancement of Artificial Intelligence ( All rights reserved. internal model of emotions with an individual optimism rate. The system is empathic in that it extracts the emotional content of the text, which is then interpreted in its own way, so the output state is not necessarily the same as the one expressed by the author of the text. The poems created by the program focus on the inspiration phrase and contain words that express those feelings. Hence the system does not just repeat the essence of the input text but expresses (in the poems) its own interpretation and emotions evoked by reading the text. As stated in (Wiggins 2012) the inspiration underlying a creative process is a spontaneous and unconscious phenomena that happens in the mind. Contrary to other problemsolving tasks, which must be performed in order, the act of creation may be executed in parallel on various levels of abstraction. This approach supports the Global Workspace Theory (Baars 1997; 2003), in which the processes in mind are performed by a group of independent specialized modules. The architecture of our system is in accordance with the assumptions of Baar s theory. We use a blackboard model consisting of a group of independent, specialized experts (the knowledge sources). In our solution, each of the experts represents some knowledge about the poetry-generation process. The experts cooperate using a common workspace - the blackboard, where all the data is stored. Various modules are triggered by the events on the blackboard. Each module, when evoked, contributes according to its knowledge, creating partial solutions, which are posted on the blackboard. These partial solutions, in turn, trigger other modules, and so on. This paper summarizes our previous work on the poetrygenerating system presented in (Misztal and Indurkhya 2014b) and (Misztal and Indurkhya 2014a) and presents a new perspective on the generation process considering its re-interpretation aspect. More details about the architecture and implementation of the system can be found in the earlier papers. This paper is organised as follows. In the Background section we describe other systems that incorporate models of emotions to interpret stimuli from the environment. We also mention applications of the blackboard-related architectures for the creative systems, and refer to the discussion about the strong and weak computational creativity. 20

2 The Overview section explains the general idea of the system. The poetry-generation process in our approach is implemented on a blackboard model, which is described in the System Architecture subsection. In this approach, the final poetry is composed by a group of experts - each of whom has some specific knowledge about the poetry-generation process, and all of them share a global workspace called the blackboard. The poetry-composition process is presented in the Poetry Generation Algorithm section. Some illustrative results are presented in the Examples section and the Evaluation contains a summary of system s performance in the context of the proposed algorithm and the final outputs evaluation. In the Conclusions section we summarize our work and propose possible continuation of this project by incorporating new experts representing other language generating systems as well as human users. Background Strong and weak computational creativity The possibility of creating a computer program which would be capable of thinking and conscious understanding has been argued during the debate over the weak vs. strong AI. One of the claims against the strong AI has been illustrated in the Chinese Room Argument in (Searle 1980), arguing that computations cannot lead to any conceptual understanding even though they provide a meaningful output. An analogous discussion has grown around the weak and strong computational creativity. In (al Rifaie and Bishop 2012) the Chinese Room experiment has been proposed in relation to creativity. The authors claim that even if the system is capable of producing syntactically correct output, it does not provide any understanding of the semantics. We agree with the general claim that the generation of emotionally-rich outputs inspired by an input implies neither understanding of the concept nor the ability to perceive emotional stimulus by the program in a human way. However, as argued by (Gervas 2010), we claim that computational modesl of conscious process may significantly differ from the original phenomenon because computers have distinct capabilities compared to humans. Incorporating cognitive models in the systems may lead to interesting solutions by simulating creative processes in the brain, but still actions performed by the computational units may be quite different from the ones going on in the brain. Systems with empathy As emotions have a strong impact on human-computer communication, there are many systems that imitate some empathic behaviors and adapt their mood to the environment. Some of them extract and replicate the emotional content of the input stimuli, while others use internal models of emotions to define the mood. In (van der Heide and Trivino 2010), the authors propose a prototype system that defines its emotional state according to changing environmental conditions according to an internal emotional model. The idea of using a model of emotions to generate emotional content in poems has been used in MASTER (Kirke and Miranda 2013), where agents in various emotional states in a society influence each other s moods with their pieces of poetry. In (Colton, Goodwin, and Veale 2012), a corpus-based poetry generator is presented that creates poems according to day s mood estimated from the news of the day. In (Davis and Mohammad 2014) the system generates pieces of music from the text according to the emotions extracted from reading the text. In our system, the emotional state is acquired based on the affective information extracted from the blog text, but it is also dependent on the individual features of the system the model of emotions and its optimism rate that give the system an individual personality. Hence the external factors are used only as an inspiration for the theme and stimulus for the affective state. Blackboard architecture in computational creativity systems In the blackboard model, the group of experts representing diverse knowledge sources cooperates using a common workspace the blackboard. It allows a range of different experts represented as diverse computational models. This architecture supports the Global Workspace Theory of mind proposed in (Baars 1997; 2003), in which the consciousness is compared to a theater where the actions are performed by a large number of autonomous specialized modules (the actors). Therefore the blackboard model has a potential to be used for simulating cognitive processes, and it has already been applied in the field of computational creativity. For example, in the Slant system for story generation (Montfort et al. 2013) the blackboard model combines separate subsystems dealing with plot, figuration, and the narrative discourse tasks in the generation problem. In (Wiggins 2012), the Global Workspace Theory has been proposed as a model to simulate creative musics composition. Another multi-agent approach to poetry generation is used in the later version of WASP (Gervas 2010), where specialized families of experts cooperate during the poetry-generation process. Groups of agents work there as a cooperative society of readers/critics/editors/writers. However, WASP does not incorporate the blackboard model directly. Overview The system structure is based on the blackboard model. It consists of a group of experts that represent diverse sources of knowledge, the common blackboard workspace and the control component that regulates the process. The modules are described in the System architecture subsection. At the beginning of the poetry-generation process, the input text is placed on the blackboard and the agents start to work on it. Each agent has a special role and knowledge and it waits until it finds something on the blackboard that it can use for performing its task. When something interesting appears, the agent processes this information using its individual knowledge and adds new partial solution to the blackboard. The control module decides which agent s contribution should be used for the final poem. The algorithm 21

3 is explained in more details in Poetry generation algorithm subsection. System architecture The system architecture is presented in Figure 1. The main modules of the system are described below. Figure 1: Blackboard architecture used in the system. The text is input to the system. A group of specialized experts processes it some of them extract information and the others generate new ideas inspired by information on the blackboard. All partial solutions are added to the common workspace until the poem is finished. The control component regulates the process by scheduling experts actions. Blackboard is a common workspace, shared by experts, with partial solutions and other information about the problem on various levels of abstraction. Detailed description of the blackboard sections is presented in our former work (Misztal and Indurkhya 2014b). Below we present a brief description of the main sections of the blackboard. Initial data Information input manually at the beginning of the process, containing the input text and the stylistic constraints such as number of lines, syllables, rhymes pattern as well as grammatical requirements concerning tense and person. Text analysis data Information extracted from the text. It consists of key phrases rated by the inspiration value (estimated as the number of associations that may be derived by the experts), the topic of the poem selected as the most inspiring phrase and the emotional state defined according to the sentiments in the text. Pool of ideas Words and partial solutions produced by the experts. The expressions are derived from the topic phrase and extended with the use of experts knowledge such as lexical resources. The words in the pool are divided into categories based on their grammatical form and meaning. They may be used as an inspiration for other experts to produce new phrases. Solutions are rated by the selection experts according to the heuristics concerning the current constraints for poem lines. Poem draft Current version of the poem consisting of lines. Each line is selected from among the phrase candidates by the evaluation experts. Model of emotions A two-dimensional model, where each emotional state is represented by coordinates in (valence, arousal) space. The emotions used in the model are WordNet hyponyms of the word emotion in the WordNet- Affect lexicon in the hierarchy of emotional categories. The (valence, arousal) coordinates for emotional labels in the model have been retrieved from the ANEW database (Bradley and Lang 2010). The emotion model is described in more details in (Misztal and Indurkhya 2014b). Experts Independent modules having access to the common blackboard. They are triggered by events on the blackboard when they find something that they can use, they add new information to the blackboard. Each module has an individual knowledge and they all play diverse roles in the system. Different experts work on the blackboard data at different abstraction levels as shown in Table 1. Control component the unit responsible for setting the initial constraints for the poem, setting the experts probabilities and selecting the expert whose contribution should be used for the current line of poem. In the current version of the system, it is possible to select the number of syllables in lines, grammar form and rhyme pattern. The importance factors for the experts are chosen manually, and are used during the generation process when an expert produces a number of phrases proportional to its importance factor. The control module also aims to maximize the diversity of the poem by giving preference to the artifacts generated by those experts that contributed less frequently before. Poetry generation algorithm We present below the general algorithm for poetry generation. (For more details with an illustrative example see (Misztal and Indurkhya 2014b).) 1. Text and constraints are input by the user. 2. Steps 2-6 are repeated until the required number of lines has been produced. 3. Text-analysis experts extract information from the input. (a) Key phrases are extracted. (b) Poem topic is selected. (c) Emotional state is defined. 4. Word-generation experts add words associated with the topic. 5. Poem-making experts construct expressions from words on the blackboard. 6. Selection experts rate phrases according to constraints for current line. 7. Control component makes the final selection of the poem line considering the experts past frequencies. Examples We present below some example outputs of the system inspired by three input texts along with our remarks about the interpretation. 22

4 Abstraction level Expert Produced artifacts/examples Text analysis level Keywords expert Key phrases Emotional Emotional state Keywords expert Words Words level WordNet expert Words - synonyms Emotional expert Emotional words Collocations level WordNet expert Hypernyms, antonyms Collocation expert Epithets, verbs, comparisons - words collocations Apostrophe expert phrases: O life the musical being Comparison expert phrases: As whole as a community Epithet expert Phrases: Free modern theatre Phrases level experts Metaphor expert Phrases: I am like the pleasant house Oxymoron expert Phrases: Huge or little Rhetorical expert Phrases: What is the royal house? Emotional expert Phrases: The tumult makes you anxious Exclamation expert Phrases: O theatre the free edifice! Repetition expert Phrases - repetition Syllables expert Selected phrases: number of syllables in lines Control level experts Rhymes expert Selected phrases: rhyme measure in lines Emotional expert Selected phrases: emotional state of the poem Random expert Selected phrase: random selection Table 1: Experts in the poetry generation process grouped by the abstraction level on which they work. Compassionate poem about the way Inspired by the text: Lately everyone has been wondering Is Jenelle and Gary going to get back together?!. NO! He is living his life and I m living mine. We are both happy with our lives the way they are at the moment, I know for me at least I m EXTREMELY happy. Gary might of been tweeting things because he might of been jealous in a way that I was dating Courtland but he agrees to stop today. Topic: Whole way, Emotion: compassion, Poem: the grand manner born tenderly oh way the cheerful property she was like the general mode guileless continuous manner as whole as a distribution she loved the mercifulness! she was like the pastel-like style Remarks: This poem has been generated from the phrase whole way that was found to be the most inspiring as it provided a wide range of word associations. Hence, the output poem is lexically diversified. The emotional state is expressed by the expression mercifulness. Anxious poem about the theater Inspired by the text: The show with KFrog out in Temecula California was incredible. It was in a huge theater at the Pechanga Casino. The show was for KFrog Cares which raises money for Loma Linda University Childrens Hospital. It was us with Katie Arminger and Jana Kramer was supposed to headline, but she got sick. So that meant we were the headliners! And that crowd could not have been more gracious, rowdy, or amazing! We had the BEST show. And we REALLY needed that. It was crazy because we sold out of ALL of the merch we brought! T- shirts and Eps flew out the door! That was an amazing feeling. Topic: Huge theater, Emotion: anxiety, Poem: great big theater vast like a majority why is the theater so entire? she needs the insecurity Remarks: For this poem the rhyme constraints were selected. Calm poem about New York Inspired by the text: Daniel and I felt so, so very fortunate to go here. The timing could not have been more perfect for us as our last few days in New York had been quite hectic. We had been trying to wrap our heads around our lives without anymore more fertility treatments. Topic: New York, Emotion: Calmness Poem: she loved the peaceable new york new like a revision she was like the york 23

5 she needed the easy house of york! Remarks: This example illustrates how the cooperation of experts within a blackboard system may lead to creative solutions. The well-known name New York has been broken up into words and processed separately. The analysis on various abstraction levels provides an original and uncommon interpretation of the topic phrase. Evaluation As presented in the Examples section, the system is capable of re-conceptualizing input text and producing artifacts that express its own interpretation of the analyzed data. As shown in the last example, use of diverse experts within the blackboard model supports creating novel and original associations at various levels of abstraction from well-known expressions. Use of emotional words creates the impression of intentionality and individual perception of the world. The optimism rate affects the emotional perception and allows a more individual interpretation. For instance, the same input may cause the anxiety or anger emotional state depending on the optimism rate. However, the input text in the system is used rather as an emotional and thematic inspiration than a concept to be reinterpreted. As only one phrase from the text is selected as the main topic for the poem, the overall sense and the context of the input may be lost in the final artifact. This aspect might be changed by taking under consideration more ideas from the text and analyzing them in parallel to produce a poem that combines more aspects of the inspiring stimulus. In our recent work (Misztal and Indurkhya 2014b), we presented an evaluation of our system considering both the results, and the process by which they were generated according to some evaluation models for computational creativity systems and poetry generators. The blackboard system provided with a model of emotions generally was capable of producing outputs satisfying the triple constraints on grammar, meaningfulness and poeticness provided by (Manurung, Ritchie, and Thompson 2012) as basic criteria for evaluating poetry produced by a computer. According to the FACE descriptive model (Colton, Charnley, and Pease 2011) our system performed creative acts of form < A g, C g, E g >. This means that the system (concept) produces expressions using a fitness function evaluating the (concept, expression) pairs with real-number values. However, it was found that a weak point of the system, according to both evaluation approaches, is the insufficient definition of constraints for the outputs. In this work, we managed to add a new constraint for the rhymes. As presented in Examples section, the system is capable of producing rhyming lines by selecting highest-rated phrases according to the constraints. However, this approach does not yet provide satisfactory results as in many cases the phrasespace does not contain any rhyming candidates. This issue needs to be further addressed and improved in the future work. Conclusions We proposed a system that is capable of re-conceptualizing emotionally rich text by emphatic interpretation and composition of poems that express feelings evoked by reading the text. The emotional state is defined by the model of emotions, and the mood is modulated by the optimism-rate factor given to the character. The blackboard architecture used in the system supports flexibility and allows for a variety of solutions on various abstraction levels leading to potentially creative solutions, thereby simulating the inspiration and invention processes in the mind. Analogous to the Baars Theater of Consciousness, the experts act on the scene of the blackboard, and their cooperative play leads to the final solution. The model also provides a way to combine diverse existing systems so that they may cooperate in the common workspace. As the continuation of our work we would like to incorporate in the system other poetry- and languagegenerating tools. We believe that the blackboard architecture has a potential to contribute to the field of computational creativity by combining many of the existing approaches in this domain. Another interesting extension of this project would be creating an interactive blackboard system that would allow human experts to participate in the generation process and add their phrases to the blackboard.the program could also visualize the generation process by displaying some of the data appearing on the blackboard during the experts work. This idea could provide a better understanding and interest in the domain of creativity. References al Rifaie, M., and Bishop, M Weak vs. strong computational creativity. In Computing, Philosophy and the Question of Bio-Machine Hybrids: 5th AISB Symposium on Computing and Philosophy, University of Birmingham, UK. Baars, B. J In the theater of consciousness. Journal of Consciousness Studies 4. Baars, B. J The global brainweb: An update on global workspace theory. Science and Consciousness Review. Bradley, M. M., and Lang, P. J Affective norms for english words(anew): Affective ratings of words and instruction manual. Technical report, NIMH Center for the Study of Emotion and Attention University of Florida. Colton, S.; Charnley, J.; and Pease, A Computational creativity theory: The face and idea descriptive models. In Proceedings of the International Conference on Computational Creativity. Colton, S.; Goodwin, J.; and Veale, T Full-face poetry generation. In Proceedings of the International Conference on Computational Creativity. Davis, H., and Mohammad, S. M Generating music from literature. In Proceedings of the EACL Workshop on Computational Linguistics for Literature, April 2014, Gothenburg, Sweden. 24

6 Gervas, P Engineering linguistic creativity: Bird flight and jet planes. In Proceedings of Workshop on Computational Approaches to Linguistic Creativity. Kirke, A., and Miranda, E Emotional and multi-agent systems in computer-aided writing and poetry. In Proceedings of the Artificial Intelligence and Poetry Symposium. Manurung, R.; Ritchie, G.; and Thompson, H Using genetic algorithms to create meaningful poetic text. Journal of Experimental and Theoretical Artificial Intelligence. Misztal, J., and Indurkhya, B. 2014a. A blackboard architecture for automatic poetry generation. In Proceedings of the Workshop Computational Creativity, Concept Invention, and General Intelligence at 21st European Conference on Artificial Intelligence. Misztal, J., and Indurkhya, B. 2014b. Poetry generation system with an emotional personality. In Proc. of the 5th International Conference on Computational Creativity. Montfort, N.; y Prez, R. P.; Fox, D. H.; and Campana, A Slant: A blackboard system to generate plot, figuration, and narrative discourse aspects of stories. In Proceedings of the Fourth International Conference on Computational Creativity. Searle, J Minds, brains, and programs. van der Heide, A., and Trivino, G Simulating emotional personality in human computer interfaces. In Proceedings of IEEE International Conference on the Fuzzy Systems (FUZZ). Wiggins, G. A The minds chorus: Creativity before consciousness. 25