We present a set of novel neural supervised and unsupervised approaches for determining the readability of documents. In the unsupervised setting, we leverage neural language models, whereas in the ...supervised setting, three different neural classification architectures are tested. We show that the proposed neural unsupervised approach is robust, transferable across languages, and allows adaptation to a specific readability task and data set. By systematic comparison of several neural architectures on a number of benchmark and new labeled readability data sets in two languages, this study also offers a comprehensive analysis of different neural approaches to readability classification. We expose their strengths and weaknesses, compare their performance to current state-of-the-art classification approaches to readability, which in most cases still rely on extensive feature engineering, and propose possibilities for improvements.
Background:
Advances in machine learning (ML) technology have opened new avenues for detection and monitoring of cognitive decline. In this study, a multimodal approach to Alzheimer's dementia ...detection based on the patient's spontaneous speech is presented. This approach was tested on a standard, publicly available Alzheimer's speech dataset for comparability. The data comprise voice samples from 156 participants (1:1 ratio of Alzheimer's to control), matched by age and gender.
Materials and Methods:
A recently developed Active Data Representation (ADR) technique for voice processing was employed as a framework for fusion of acoustic and textual features at sentence and word level. Temporal aspects of textual features were investigated in conjunction with acoustic features in order to shed light on the temporal interplay between paralinguistic (acoustic) and linguistic (textual) aspects of Alzheimer's speech. Combinations between several configurations of ADR features and more traditional bag-of-n-grams approaches were used in an ensemble of classifiers built and evaluated on a standardised dataset containing recorded speech of scene descriptions and textual transcripts.
Results:
Employing only semantic bag-of-n-grams features, an accuracy of 89.58% was achieved in distinguishing between Alzheimer's patients and healthy controls. Adding temporal and structural information by combining bag-of-n-grams features with ADR audio/textual features, the accuracy could be improved to 91.67% on the test set. An accuracy of 93.75% was achieved through late fusion of the three best feature configurations, which corresponds to a 4.7% improvement over the best result reported in the literature for this dataset.
Conclusion:
The proposed combination of ADR audio and textual features is capable of successfully modelling temporal aspects of the data. The machine learning approach toward dementia detection achieves best performance when ADR features are combined with strong semantic bag-of-n-grams features. This combination leads to state-of-the-art performance on the AD classification task.
Learning from texts has been widely adopted throughout industry and science. While state-of-the-art neural language models have shown very promising results for text classification, they are ...expensive to (pre-)train, require large amounts of data and tuning of hundreds of millions or more parameters. This paper explores how automatically evolved text representations can serve as a basis for explainable, low-resource branch of models with competitive performance that are subject to automated hyperparameter tuning. We present autoBOT (automatic Bags-Of-Tokens), an autoML approach suitable for low resource learning scenarios, where both the hardware and the amount of data required for training are limited. The proposed approach consists of an evolutionary algorithm that jointly optimizes various sparse representations of a given text (including word, subword, POS tag, keyword-based, knowledge graph-based and relational features) and two types of document embeddings (non-sparse representations). The key idea of autoBOT is that, instead of evolving at the learner level, evolution is conducted at the representation level. The proposed method offers competitive classification performance on fourteen real-world classification tasks when compared against a competitive autoML approach that evolves ensemble models, as well as state-of-the-art neural language models such as BERT and RoBERTa. Moreover, the approach is explainable, as the importance of the parts of the input space is part of the final solution yielded by the proposed optimization procedure, offering potential for meta-transfer learning.
In this paper, we address the task of zero-shot cross-lingual news sentiment classification. Given the annotated dataset of positive, neutral, and negative news in Slovene, the aim is to develop a ...news classification system that assigns the sentiment category not only to Slovene news, but to news in another language without any training data required. Our system is based on the multilingual BERTmodel, while we test different approaches for handling long documents and propose a novel technique for sentiment enrichment of the BERT model as an intermediate training step. With the proposed approach, we achieve state-of-the-art performance on the sentiment analysis task on Slovenian news. We evaluate the zero-shot cross-lingual capabilities of our system on a novel news sentiment test set in Croatian. The results show that the cross-lingual approach also largely outperforms the majority classifier, as well as all settings without sentiment enrichment in pre-training.
In this paper, we look at the issue of reproducibility and replicability in bilingual terminology alignment (BTA). We propose a set of best practices for reproducibility and replicability of NLP ...papers and analyze several influential BTA papers from this perspective. Next, we present our attempts at replication and reproduction, where we focus on a bilingual terminology alignment approach described by Aker et al. (Extracting bilingual terminologies from comparable corpora. In: Proceedings of the 51st annual meeting of the association for computational linguistics, vol. 1 402–411,
2013
) who treat bilingual term alignment as a binary classification problem and train an SVM classifier on various dictionary and cognate-based features. Despite closely following the original paper with only minor deviations—in areas where the original description is not clear enough—we obtained significantly worse results than the authors of the original paper. We then analyze the reasons for the discrepancy and describe our attempts at adaptation of the approach to improve the results. Only after several adaptations, we achieve results which are close to the results published in the original paper. Finally, we perform the experiments to verify the replicability and reproducibility of our own code. We publish our code and datasets online to assure the reproducibility of the results of our experiments and implement the selected BTA models in an online platform making them easily reusable even by the technically less-skilled researchers.
Platforms that feature user-generated content (social media, online forums, newspaper comment sections etc.) have to detect and filter offensive speech within large, fast-changing datasets. While ...many automatic methods have been proposed and achieve good accuracies, most of these focus on the English language, and are hard to apply directly to languages in which few labeled datasets exist. Recent work has therefore investigated the use of cross-lingual transfer learning to solve this problem, training a model in a well-resourced language and transferring to a less-resourced target language; but performance has so far been significantly less impressive. In this paper, we investigate the reasons for this performance drop, via a systematic comparison of pre-trained models and intermediate training regimes on five different languages. We show that using a better pre-trained language model results in a large gain in overall performance and in zero-shot transfer, and that intermediate training on other languages is effective when little target-language data is available. We then use multiple analyses of classifier confidence and language model vocabulary to shed light on exactly where these gains come from and gain insight into the sources of the most typical mistakes.
Predstavljamo metodo za luščenje definicij iz specializiranih korpusov. Metoda je bila razvita za slovenščino in angleščino, sestavljajo pa jo trije pristopi: v prvem definicije luščimo z ...leksikoskladenjskimi vzorci, drugi uporablja avtomatsko izluščeno terminologijo, tretji pa lušči stavke, v katerih se nahajata pojem in njegova nadpomenka iz semantičnega leksikona wordnet. Metodologijo smo preizkusili na primeru področja jezikovnih tehnologij. Za namene modeliranja izbranega področja smo zgradili primerljivi slovensko-angleški Korpus jezikovnih tehnologij, izluščene kandidate pa smo uporabili pri gradnji Glosarja jezikovnih tehnologij. Celotno metodologijo smo strnili v prosto dostopen delotok, implementiran v spletnem okolju za gradnjo delotokov Clowdflows. V delotok lahko uporabnik prek spleta naloži korpus v različnih formatih, ga jezikoslovno označi, izlušči terminologijo in kandidate za definicije ter rezultate vizualizira ali shrani.
In recent years, the use of deep neural networks and dense vector embeddings for text representation have led to excellent results in the field of computational understanding of natural language. It ...has also been shown that word embeddings often capture gender, racial and other types of bias. The article focuses on evaluating Slovene and Croatian word embeddings in terms of gender bias using word analogy calculations. We compiled a list of masculine and feminine nouns for occupations in Slovene and evaluated the gender bias of fastText, word2vec and ELMo embeddings with different configurations and different approaches to analogy calculations. The lowest occupational gender bias was observed with the fastText embeddings. Similarly, we compared different fastText embeddings on Croatian occupational analogies.
This paper presents a novel neural architecture capable of outperforming state-of-the-art systems on the task of language variety classification. The architecture is a hybrid that combines ...character-based convolutional neural network (CNN) features with weighted bag-of-n-grams (BON) features and is therefore capable of leveraging both character-level and document/corpus-level information. We tested the system on the Discriminating between Similar Languages (DSL) language variety benchmark data set from the VarDial 2017 DSL shared task, which contains data from six different language groups, as well as on two smaller data sets (the Arabic Dialect Identification (ADI) Corpus and the German Dialect Identification (GDI) Corpus, from the VarDial 2016 ADI and VarDial 2018 GDI shared tasks, respectively). We managed to outperform the winning system in the DSL shared task by a margin of about 0.4 percentage points and the winning system in the ADI shared task by a margin of about 0.2 percentage points in terms of weighted F1 score without conducting any language group-specific parameter tweaking. An ablation study suggests that weighted BON features contribute more to the overall performance of the system than the CNN-based features, which partially explains the uncompetitiveness of deep learning approaches in the past VarDial DSL shared tasks. Finally, we have implemented our system in a workflow, available in the ClowdFlows platform, in order to make it easily available also to the non-programming members of the research community.
•Automatic feature selection methods improve emotion recognition performance.•Systematic evaluation of four feature selection methods is performed on three datasets in three languages and two ...acoustic feature sets.•Our recently introduced Active feature selection method is successfully tested on the emotion recognition task.•With only 1/3 of eGeMAPs and 1/10 emobase features comparable results to the full feature set are obtained.
Research in automatic affect recognition has seldom addressed the issue of computational resource utilization. With the advent of ambient intelligence technology which employs a variety of low-power, resource-constrained devices, this issue is increasingly gaining interest. This is especially the case in the context of health and elderly care technologies, where interventions may rely on monitoring of emotional status to provide support or alert carers as appropriate. This paper focuses on emotion recognition from speech data, in settings where it is desirable to minimize memory and computational requirements. Reducing the number of features for inductive inference is a route towards this goal. In this study, we evaluate three different state-of-the-art feature selection methods: Infinite Latent Feature Selection (ILFS), ReliefF and Fisher (generalized Fisher score), and compare them to our recently proposed feature selection method named ‘Active Feature Selection’ (AFS). The evaluation is performed on three emotion recognition data sets (EmoDB, SAVEE and EMOVO) using two standard acoustic paralinguistic feature sets (i.e. eGeMAPs and emobase). The results show that similar or better accuracy can be achieved using subsets of features substantially smaller than the entire feature set. A machine learning model trained on a smaller feature set will reduce the memory and computational resources of an emotion recognition system which can result in lowering the barriers for use of health monitoring technology.