Humor is a fundamental aspect of human communication and cognition, as it plays a crucial role in social engagement. Although theories about humor have evolved over centuries, there is still no agreement on a single, comprehensive humor theory. Likewise, computationally recognizing humor remains a significant challenge despite recent advances in large language models. Moreover, most computational approaches to detecting humor are not based on existing humor theories. This paper contributes to bridging this long-standing gap between humor theory research and computational humor detection by creating an interpretable framework for humor classification, grounded in multiple humor theories, called THInC (Theory-driven Humor Interpretation and Classification). THInC ensembles interpretable GA2M classifiers, each representing a different humor theory. We engineered a transparent flow to actively create proxy features that quantitatively reflect different aspects of theories. An implementation of this framework achieves an F1 score of 0.85. The associative interpretability of the framework enables analysis of proxy efficacy, alignment of joke features with theories, and identification of globally contributing features. This paper marks a pioneering effort in creating a humor detection framework that is informed by diverse humor theories and offers a foundation for future advancements in theory-driven humor classification. It also serves as a first step in automatically comparing humor theories in a quantitative manner.

Recent advances in neural-symbolic learning, such as DeepProbLog, extend probabilistic logic programs with neural predicates. Like graphical models, these probabilistic logic programs define a probability distribution over possible worlds, for which inference is computationally hard. We propose DeepStochLog, an alternative neural-symbolic framework based on stochastic definite clause grammars, a kind of stochastic logic program. More specifically, we introduce neural grammar rules into stochastic definite clause grammars to create a framework that can be trained end-to-end. We show that inference and learning in neural stochastic logic programming scale much better than for neural probabilistic logic programs. Furthermore, the experimental evaluation shows that DeepStochLog achieves state-of-the-art results on challenging neural-symbolic learning tasks.

Large transformer-based language models, e.g. BERT and GPT-3, outperform previous architectures on most natural language processing tasks. Such language models are first pre-trained on gigantic corpora of text and later used as base-model for finetuning on a particular task. Since the pre-training step is usually not repeated, base models are not up-to-date with the latest information. In this paper, we update RobBERT, a RoBERTa-based state-of-the-art Dutch language model, which was trained in 2019. First, the tokenizer of RobBERT is updated to include new high-frequent tokens present in the latest Dutch OSCAR corpus, e.g. corona-related words. Then we further pre-train the RobBERT model using this dataset. To evaluate if our new model is a plug-in replacement for RobBERT, we introduce two additional criteria based on concept drift of existing tokens and alignment for novel tokens.We found that for certain language tasks this update results in a significant performance increase. These results highlight the benefit of continually updating a language model to account for evolving language use.

Pre-trained large-scale language models such as BERT have gained a lot of attention thanks to their outstanding performance on a wide range of natural language tasks. However, due to their large number of parameters, they are resource-intensive both to deploy and to fine-tune. Researchers have created several methods for distilling language models into smaller ones to increase efficiency, with a small performance trade-off. In this paper, we create several different distilled versions of the state-of-the-art Dutch RobBERT model and call them RobBERTje. The distillations differ in their distillation corpus, namely whether or not they are shuffled and whether they are merged with subsequent sentences. We found that the performance of the models using the shuffled versus non-shuffled datasets is similar for most tasks and that randomly merging subsequent sentences in a corpus creates models that train faster and perform better on tasks with long sequences. Upon comparing distillation architectures, we found that the larger DistilBERT architecture worked significantly better than the Bort hyperparametrization. Interestingly, we also found that the distilled models exhibit less gender-stereotypical bias than its teacher model. Since smaller architectures decrease the time to fine-tune, these models allow for more efficient training and more lightweight deployment of many Dutch downstream language tasks.

Large pre-trained transformer-based language models have revolutionized the field of natural language processing in recent years. While BERT-like models perform exceptionally well for analytical tasks such as classification and regression, their text generation capabilities are usually limited to predicting tokens within a given context. In this paper, we introduce G ALMET , a model that generates text by using genetic algorithms with BERT-like language models for evolving text. We use G ALMET with the RoBERTa language model to automatically evolve real headlines into more satirical headlines. This is achieved by adapting the masked language head to the headlines domain for the mutation operator and finetuning a regression head to distinguish headlines from satire for the fitness function. We evaluated our system by comparing generated satirical head-lines against human-edited headlines and just the fine-tuned masked language head. We found that while humans generally outperform the model, generations by G ALMET are also often preferred over human-edited headlines. However, we also found that only using the fine-tuned masked language model gives slightly preferred satire due to generating more readable sentences. G ALMET is thus a first step towards a new way of creating text generators using masked language models by transforming text guided by scores from another language model.

Slideshows are a popular tool for presenting information in a structured and attractive manner. There exists a wide range of different slideshows editors, often with their own proprietary encoding that is incompatible with other editors. Merging slideshows from different editors and making the slide design consistent is a nontrivial and time-intensive task. We introduce SandSlide, the first system for automatically normalizing a deck of slides from a PDF file into an editable PowerPoint file that adheres to the default slide templates, and is thus able to fully leverage the flexible layout capabilities of modern slideshow editors. SandSlide achieves this by labeling objects, using a qualitative representation to find the most similar slide layout and aligning content from the slide with this layout. To evaluate SandSlide, we collected and annotated slides from different slideshows. Our experiments show that a greedy search is able to obtain high responsiveness on supported and almost supported slides, and that a significant majority of slides fall into this category. Additionally, our annotated dataset contains fine-grained annotations on different properties of slideshows to further incentivize research on all aspects of the problem of slide normalization.

Can a computer have a sense of humor?On reflection, this question may seem paradoxical given that humor is such an intrinsically human trait.While some primates produce laughter (Gervais & Wilson, 2005;Preuschoft & van Hooff, 1997), humans are the only known species that use humor for making others laugh.Furthermore, every known human civilization also has had at least some form of humor for making others laugh (Caron, 2002;Gervais & Wilson, 2005).Given this innately and intrinsically human skill, it might seem implausible for a machine to have any form of a sense of humor.In this article, we explore several computational techniques that have been used to detect and generate humor, and how these systems can help us improve our humor abilities and gain more insights into this uniquely human trait.

Designers increasingly rely on procedural generation for automatic generation of content in various industries. These techniques require extensive knowledge of the desired content, and about how to actually implement such procedural methods. Algorithms for learning interpretable generative models from example content could alleviate both difficulties. We propose SIGI, a novel method for inferring shapes and inducing a shape grammar from grid-based 3D building examples. This interpretable grammar is well-suited for co-creative design. Applied to Minecraft buildings, we show how the shape grammar can be used to automatically generate new buildings in a similar style.

Portmanteaus are a type of neologism combining two source words, for example brunch (from breakfast and lunch), and are popular for naming all kinds of phenomena. While coming up with suitable portmanteaus is a difficult creative endeavor, several portmanteau generators already exist for assistance in this process. When using these systems, it is often hard to find out which of the generated portmanteau is likely to be the best, and consequently also hard to automatically compare the quality of different portmanteau generators. In this paper, we create a model that can rank portmanteaus for two given source words, which thus aims to help find the best portmanteau to help further improve portmanteau generators. Our model first uses XGBoost trained on unlabeled generated outputs and existing portmanteaus to learn to rank portmanteaus and shows that this already greatly improves the performance of the initial generator. By ranking outputs of a state-of-the-art generator and a new simple portmanteau generator, we show by validating its quality in a human evaluation that the ranker can help visually identify the better generator, thus providing an alternative to only calculating real portmanteau generation frequency. Additionally, we find that this first model performs almost as well as a model trained on more fine-grained human-labeled portmanteaus. This indicates that just using generated and real portmanteaus is enough to create a ranker that can in turn improve the quality of the initial generator, and could additionally be of use in comparing different portmanteau generators.

Rosetta Code provides improv theatre performers with artificial intelligence (AI)-based technology to perform shows understandable across many different languages. We combine speech recognition, improv chatbots and language translation tools to enable improvisers to communicate with each other while being understood—or comically misunderstood—by multilingual audiences. We describe the technology underlying Rosetta Code, detailing the speech recognition, machine translation, text generation and text-to-speech subsystems. We then describe scene structures that feature the system in performances in multilingual shows (9 languages). We provide evaluative feedback from performers, audiences, and critics. From this feedback, we draw analogies between surrealism, absurdism, and multilingual AI improv. Rosetta Code creates a new form of language-based absurdist improv. The performance remains ephemeral and performers of different languages can express themselves and their culture while accommodating the linguistic diversity of audiences.

Natural language generation provides designers with methods for automatically generating text, e.g. for creating summaries, chatbots and game content. In practise, text generators are often either learned and hard to interpret, or created by hand using techniques such as grammars and templates. In this paper, we introduce a novel grammar induction algorithm for learning interpretable grammars for generative purposes, called Gitta. We also introduce the novel notion of template trees to discover latent templates in corpora to derive these generative grammars. By using existing human-created grammars, we found that the algorithm can reasonably approximate these grammars using only a few examples. These results indicate that Gitta could be used to automatically learn interpretable and easily modifiable grammars, and thus provide a stepping stone for human-machine co-creation of generative models.

Detecting if a text is humorous is a hard task to do computationally, as it usually requires linguistic and common sense insights. In machine learning, humor detection is usually modeled as a binary classification task, trained to predict if the given text is a joke or another type of text. Rather than using completely different non-humorous texts, we propose using text generation algorithms for imitating the original joke dataset to increase the difficulty for the learning algorithm. We constructed several different joke and non-joke datasets to test the humor detection abilities of different language technologies. In particular, we compare the humor detection capabilities of classic neural network approaches with the state-of-the-art Dutch language model RobBERT. In doing so, we create and compare the first Dutch humor detection systems. We found that while other language models perform well when the non-jokes came from completely different domains, RobBERT was the only one that was able to distinguish jokes from generated negative examples. This performance illustrates the usefulness of using text generation to create negative datasets for humor recognition, and also shows that transformer models are a large step forward in humor detection.

Pre-trained language models have been dominating the field of natural language processing in recent years, and have led to significant performance gains for various complex natural language tasks. One of the most prominent pre-trained language models is BERT, which was released as an English as well as a multilingual version. Although multilingual BERT performs well on many tasks, recent studies show that BERT models trained on a single language significantly outperform the multilingual version. Training a Dutch BERT model thus has a lot of potential for a wide range of Dutch NLP tasks. While previous approaches have used earlier implementations of BERT to train a Dutch version of BERT, we used RoBERTa, a robustly optimized BERT approach, to train a Dutch language model called RobBERT. We measured its performance on various tasks as well as the importance of the fine-tuning dataset size. We also evaluated the importance of language-specific tokenizers and the model's fairness. We found that RobBERT improves state-of-the-art results for various tasks, and especially significantly outperforms other models when dealing with smaller datasets. These results indicate that it is a powerful pre-trained model for a large variety of Dutch language tasks. The pre-trained and fine-tuned models are publicly available to support further downstream Dutch NLP applications.

Many computer systems are becoming increasingly tailored to their users, customizing and optimizing their experience. However, most conversational agents do not follow this trend when it comes to humorous interactions. Instead, they employ pre-written answers regardless of whether the user liked previous similar interactions. While there already exist several computational humor systems that can successfully generate jokes, their joke generation models, parameters or even both are often fixed. In this paper, we propose GOOFER, a general framework for computational humor that learns joke structures and parameterizations from rated example jokes. This framework uses metrical schemas, a new notion we introduce, which are a generalization of several types of other schemas. This new type of schema makes regular schemas compatible with machine learning techniques. We also propose a strategy for identifying useful humor metrics based on humor theory, which can be used as features for the machine learning algorithm. The GOOFER framework uses these novel concepts to construct a pipeline with new components around previous generators. Using a mapping to our previous work on analogy jokes, we show that this framework cannot only generate this type of jokes well, but also find the importance of specific humor metrics for template values. This indicates that it is on the right track towards joke generation systems that can automatically learn new templates and schemas from rated examples. This work thus forms a stepping stone towards creating programs with a sense of humor that is adaptable to the user.

Talented public speakers have thousands of hours of practice. One means of improving public speaking skills is practice through improvisation, e.g. presenting an improvised presentation using an unseen slide deck. We present Tedric, a novel system capable of generating coherent slide decks based on a single topic suggestion. It combines semantic word webs with text and image data sources to create an engaging slide deck with an overarching theme. We found that audience members perceived the quality of improvised presentations using these generated slide decks to be on par with presentations using human created slide decks for the Improvised TED Talk performance format. Tedric is thus a valuable new creative tool for improvisers to perform with, and for anyone looking to improve their presentation skills.

Conversational agents, such as chatbots and virtual assistants, are typically modelled to have a broad, generic personality, which they employ in their communication with single human beings. However, by framing a conversational agent as existing fictional characters, humans can imagine a shallow agent to have a larger personality than without this framing. Using multiple such agents allows for conversational interactions that help construct stories with or without human intervention, leading to multi-agent human-computer interactive story telling. In this paper, we model six semi-independent Twitterbots based on fictional characters based on the Belgian children’s TV show Samson & Gert, which are mutually interactive with each other as well as with other Twitter users. To achieve this, we first introduce a new language for modelling generative weighted context-free grammars called Babbly and a new framework for easily specifying complex Twitterbot behaviour. We found that these bots were not only well received by users, but also created lots of interesting, unexpected positive interactions. Using fictional characters as framing for conversational agents can thus help achieving interesting personalities and shows potential in interactive computational story telling.

This paper describes the accepted entries to the seventh Playable Experiences track to be held at the 15th AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE'19). The Playable Experiences track showcases innovative complete works that are informed, inspired, or otherwise enabled by artificial intelligence.

This paper describes the accepted entries to the seventh Playable Experiences track to be held at the 15th AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE’19). The Playable Experiences track showcases innovative complete works that are informed, inspired, or otherwise enabled by artificial intelligence.

Automatically imitating input text is a common task in natural language generation, often used to create humorous results. Classic algorithms for learning to imitate text, e.g. simple Markov chains, usually have a trade-off between originality and syntactic correctness. We present two ways of automatically parodying philosophical statements from examples overcoming this issue, and show how these can work in interactive systems as well. The first algorithm uses interpolated Markov models with extensions to improve the quality of the generated texts. For the second algorithm, we propose dynamically extracting templates and filling these with new content. To illustrate these algorithms, we implemented TorfsBot, a Twitterbot imitating the witty, semi-philosophical tweets of professor Rik Torfs, the previous KU Leuven rector. We found that users preferred generative models that focused on local coherent sentences, rather than those mimicking the global structure of a philosophical statement. The proposed algorithms are thus valuable new tools for automatic parody as well as template learning systems.