Abstract In the development of a task-oriented dialogue system, defining the dialogue structure is a time-consuming task. Hence, several works have looked into automatically inferring it from data, e.g., actual conversations between a customer and a support agent. To recover such dialogue structure, recent methods based on discrete variational models learn to jointly encode and cluster utterances in dialogue states, but (i) represent utterances by only considering preceding dialogue context, and (ii) are slow to train since they are optimized with a compute-expensive decoding objective. We revisit and improve upon an existing efficient pipeline approach, commonly adopted as a baseline, that first encodes utterances and then clusters them with k-means to induce the dialogue structure. However, the existing approach represents utterances as bag-of-words or skip-thought vectors, which have been shown to perform poorly in semantic similarity tasks, and without considering dialogue context. We therefore first investigate the use of more powerful transformer-based encoders for encoding utterances. Next, we propose ellodar, a method for learning representations that capture both preceding and subsequent dialogue context, inspired by word-to-vec training strategies. ellodar is efficient since representations are learned directly in the encoding space by finetuning just a single linear layer on top of a frozen sentence encoder with a vector-to-vector regression training objective. Extensive experiments on representative datasets for dialogue structure induction (SimDial, Schema Guided Dialogues, DSTC2, and CamRest676) demonstrate that in terms of effectiveness to induce the correct dialogue structure, (i) clustering utterances represented by transformed-based encoders improves recent joint models by 13%–32% on standard cluster metrics, and (ii) clustering ellodar’s representations yields additional improvements ranging from +20% to +26%, with speedups of $$\times $$ × $$\textbf{10}$$ 10 – $$\textbf{10}^{\textbf{4}}$$ 10 4 compared to the recent joint models.

For text classification tasks, finetuned language models perform remarkably well.Yet, they tend to rely on spurious patterns in training data, thus limiting their performance on outof-distribution (OOD) test data.Among recent models aiming to avoid this spurious pattern problem, adding extra counterfactual samples to the training data has proven to be very effective.Yet, counterfactual data generation is costly since it relies on human annotation.Thus, we propose a novel solution that only requires annotation of a small fraction (e.g., 1%) of the original training data, and uses automatic generation of extra counterfactuals in an encoding vector space.We demonstrate the effectiveness of our approach in sentiment classification, using IMDb data for training and other sets for OOD tests (i.e., Amazon, SemEval and Yelp).We achieve noticeable accuracy improvements by adding only 1% manual counterfactuals: +3% compared to adding +100% in-distribution training samples, +1.3% compared to alternate counterfactual approaches.

Powerful sentence encoders trained for multiple languages are on the rise.These systems are capable of embedding a wide range of linguistic properties into vector representations.While explicit probing tasks can be used to verify the presence of specific linguistic properties, it is unclear whether the vector representations can be manipulated to indirectly steer such properties.For efficient learning, we investigate the use of a geometric mapping in embedding space to transform linguistic properties, without any tuning of the pre-trained sentence encoder or decoder.We validate our approach on three linguistic properties using a pre-trained multilingual autoencoder and analyze the results in both monolingual and crosslingual settings.

Inducing sparseness while training neural networks has been shown to yield models with a lower memory footprint but similar effectiveness to dense models.However, sparseness is typically induced starting from a dense model, and thus this advantage does not hold during training.We propose techniques to enforce sparseness upfront in recurrent sequence models for NLP applications, to also benefit training.First, in language modeling, we show how to increase hidden state sizes in recurrent layers without increasing the number of parameters, leading to more expressive models.Second, for sequence labeling, we show that word embeddings with predefined sparseness lead to similar performance as dense embeddings, at a fraction of the number of trainable parameters.

Character-level features are currently used in different neural network-based natural language processing algorithms.However, little is known about the character-level patterns those models learn.Moreover, models are often compared only quantitatively while a qualitative analysis is missing.In this paper, we investigate which character-level patterns neural networks learn and if those patterns coincide with manually-defined word segmentations and annotations.To that end, we extend the contextual decomposition (Murdoch et al., 2018) technique to convolutional neural networks which allows us to compare convolutional neural networks and bidirectional long short-term memory networks.We evaluate and compare these models for the task of morphological tagging on three morphologically different languages and show that these models implicitly discover understandable linguistic rules.

Motivation: During the last decade, improvements in high-throughput sequencing have generated a wealth of genomic data. Functionally interpreting these sequences and finding the biological signals that are hallmarks of gene function and regulation is currently mostly done using automated genome annotation platforms, which mainly rely on integrated machine learning frameworks to identify different functional sites of interest, including splice sites. Splicing is an essential step in the gene regulation process, and the correct identification of splice sites is a major cornerstone in a genome annotation system. Results: In this paper, we present SpliceRover, a predictive deep learning approach that outperforms the state-of-the-art in splice site prediction. SpliceRover uses convolutional neural networks (CNNs), which have been shown to obtain cutting edge performance on a wide variety of prediction tasks. We adapted this approach to deal with genomic sequence inputs, and show it consistently outperforms already existing approaches, with relative improvements in prediction effectiveness of up to 80.9% when measured in terms of false discovery rate. However, a major criticism of CNNs concerns their 'black box' nature, as mechanisms to obtain insight into their reasoning processes are limited. To facilitate interpretability of the SpliceRover models, we introduce an approach to visualize the biologically relevant information learnt. We show that our visualization approach is able to recover features known to be important for splice site prediction (binding motifs around the splice site, presence of polypyrimidine tracts and branch points), as well as reveal new features (e.g. several types of exclusion patterns near splice sites). Availability and implementation: SpliceRover is available as a web service. The prediction tool and instructions can be found at http://bioit2.irc.ugent.be/splicerover/. Supplementary information: Supplementary data are available at Bioinformatics online.

Broadcasters produce vast collections of video content. However, the lack of fine-grained annotations makes it difficult to retrieve video fragments of interest from these vast collections. Indeed, manual annotation of video content is labour-intensive and time-consuming. Moreover, the applicability of algorithms for automatic annotation of video content is limited, given that too many prerequisites need to be fulfilled and that a lot of concepts are unidentifiable. At the same time, people are using social media to share their thoughts about the content they view on television. Therefore, in this Ph.D. research, we plan to investigate novel machine learning-based approaches towards the task of fine-grained annotation of broadcast video content, fusing the collective knowledge present in social media with the output of audio-visual content analysis algorithms.