539 research outputs found
Sort by
Sourabrata Mukherjee: Position Paper on Stylized Dialog Response Generation
No full textMy primary research focus lies in the domain of Text Style Transfer (TST), a fascinating area within Natural Language Processing (NLP). TST involves the transformation of text into a desired style while approximately preserving its underlying content. In my research, I am also driven by the goal of incorporating TST techniques into NLP systems, particularly within the realm of dialogue systems. I am intrigued by the concept of Stylized Dialog Response Generation, which aims to enhance the versatility and adaptability of dialog systems in generating text responses with specific style attributes. By advancing our understanding of TST and its integration into dialogue systems, my research seeks to contribute to the broader field of human-computer interaction. Through the development of robust and versatile dialogue systems with enhanced style transfer capabilities, we can facilitate more engaging and personalized conversational experiences
Are Experts Needed? On Human Evaluation of Counselling Reflection Generation
No full textReflection is a crucial counselling skill where the therapist conveys to the client their interpretation of what the client said. Language models have recently been used to generate reflections automatically, but human evaluation is challenging, particularly due to the cost of hiring experts. Laypeople-based evaluation is less expensive and easier to scale, but its quality is unknown for reflections. Therefore, we explore whether laypeople can be an alternative to experts in evaluating a fundamental quality aspect: coherence and context-consistency. We do so by asking a group of laypeople and a group of experts to annotate both synthetic reflections and human reflections from actual therapists. We find that both laypeople and experts are reliable annotators and that they have moderate-to-strong inter-group correlation, which shows that laypeople can be trusted for such evaluations. We also discover that GPT-3 mostly produces coherent and consistent reflections, and we explore changes in evaluation results when the source of synthetic reflections changes to GPT-3 from the less powerful GPT-2
Better Translation + Split and Generate for Multilingual RDF-to-Text (WebNLG 2023)
No full textThis paper presents system descriptions of our submitted outputs for WebNLG Challenge 2023. We use mT5 in multi-task and multilingual settings to generate more fluent and reliable verbalizations of the given RDF triples. Furthermore, we introduce a partial decoding technique to produce more elaborate yet simplified outputs. Additionally, we demonstrate the significance of employing better translation systems in creating training data
Storyboarder
No full textA tool for the automatic generation of textual stories with accompanying images, sounds and videos.
StoryBoarder binds together text generation (GPT), image generation (StableDiffusion) and text-to-speech (Coqui) models to provide a tool for content creators to quickly generate stories. The creators have full control over the parameters of the individual tools, with the StoryBoarder binding the tools together to be easy to use while providing useful default settings and hints. The generated videos can then be directly streamed to video streaming servers (Twitch or YouTube)
Open Calls and Pilot Projects
No full textWe describe the two ELG open calls for pilot projects, the objective of
which was to demonstrate the use and the advantages of ELG in providing basic LT
for applications and as a basis for more advanced LT-based modules or components
useful to industry. Our main goal was to attract SMEs and research organisations
to either contribute additional tools or resources to the ELG platform (type A pilot
projects) or develop applications using Language Technologies available in the ELG
platform (type B pilot projects). We start with the detailed description of the submission
and evaluation processes, followed by a presentation of the open call results.
Afterwards we describe the supervision and evaluation of the execution phase of the
projects, as well as lessons learned. Overall, we were very satisfied with the setup
and with the results of the pilot projects, which demonstrate an enormous interest in
ELG and the Language Technology topic in general
Learning capabilities in Transformer Neural Networks
No full textAlthough the contemporary neural networks, inspired by biological neurons, were able to reach human-like performance on many tasks in recent years, their optimization (learning) process is still very far from the one observed in humans.
This thesis investigates various aspects of learning in the current state-of-the-art Transformer neural networks, the dominant architecture in the current neural language processing.
Firstly, we measure the level of generalization in Transformers using several probing experiments based on the idea of adversarial evaluation. Secondly, we explore their potential for incremental learning when combined with regularization using the elastic weight consolidation approach.
Lastly, we propose a modular extension of the existing Transformer architecture enabling subnetwork selection conditioned on the intermediate hidden layer outputs and analyze the attributes of this network modularization.
We investigate our hypotheses mainly within the scope of neural machine translation and multilingual translation showing the limitations of the original Transformer and the elastic weights consolidation regularization while presenting promising results of the novel modular Transformer architecture
UFAL-ULD at BLP-2023 Task 2 Sentiment Classification in Bangla Text
No full textIn this paper, we present the UFAL-ULD team's system for the BLP Shared Task 2: Sentiment Analysis of Bangla Social Media Posts. The Task 2 involves classifying text into Positive, Negative, or Neutral sentiments. As a part of this task, we conducted a series of experiments with several pre-trained sequence classification models -- XLM-RoBERTa, BanglaBERT, Bangla BERT Base and Multilingual BERT. Among these, our best-performing model was based on the XLM-RoBERTa-base architecture, which outperforms baseline models. Our system was ranked 19th among the 30 teams that participated in the task
Generating clickbait spoilers with an ensemble of large language models
No full textClickbait posts are a widespread problem in the webspace. The generation of spoilers, i.e. short texts that neutralize clickbait by providing information that makes it uninteresting, is one of the proposed solutions to the problem. Current state-of-the-art methods are based on passage retrieval or question answering approaches and are limited to generating spoilers only in the form of a phrase or a passage. In this work, we propose an ensemble of fine-tuned large language models for clickbait spoiler generation. Our approach is not limited to phrase or passage spoilers, but is also able to generate multipart spoilers that refer to several non-consecutive parts of text. Experimental evaluation demonstrates that the proposed ensemble model outperforms the baselines in terms of BLEU, METEOR and BERTScore metrics
Mind the Labels: Describing Relations in Knowledge Graphs With Pretrained Models
No full textPretrained language models (PLMs) for data-to-text (D2T) generation can use human-readable data labels such as column headings, keys, or relation names to generalize to out-of-domain examples. However, the models are well-known in producing semantically inaccurate outputs if these labels are ambiguous or incomplete, which is often the case in D2T datasets. In this paper, we expose this issue on the task of descibing a relation between two entities. For our experiments, we collect a novel dataset for verbalizing a diverse set of 1,522 unique relations from three large-scale knowledge graphs (Wikidata, DBPedia, YAGO). We find that although PLMs for D2T generation expectedly fail on unclear cases, models trained with a large variety of relation labels are surprisingly robust in verbalizing novel, unseen relations. We argue that using data with a diverse set of clear and meaningful labels is key to training D2T generation systems capable of generalizing to novel domains
Velké jazykové modely: Co znamená velké a co jazykové?
No full textThe talk introduced large language models: their training and application, and related research conducted at ÚFAL MFF UK