arXiv subjects

Kairit Sirts

Publications and source records attributed to Kairit Sirts.

29 records · Page 2Linked to original sources

arXiv2021

EstBERT: A Pretrained Language-Specific BERT for Estonian

This paper presents EstBERT, a large pretrained transformer-based language-specific BERT model for Estonian. Recent work has evaluated multilingual BERT models on Estonian tasks and found them to outperform the baselines. Still, based on existing studies on other languages, a language-specific BERT model is expected to improve over the multilingual ones. We first describe the EstBERT pretraining process and then present the results of the models based on finetuned EstBERT for multiple NLP tasks, including POS and morphological tagging, named entity recognition and text classification. The evaluation results show that the models based on EstBERT outperform multilingual BERT models on five tasks out of six, providing further evidence towards a view that training language-specific BERT models are still useful, even when multilingual models are available.

cs.CL↗

arXiv2021

Enhancing Sequence-to-Sequence Neural Lemmatization with External Resources

We propose a novel hybrid approach to lemmatization that enhances the seq2seq neural model with additional lemmas extracted from an external lexicon or a rule-based system. During training, the enhanced lemmatizer learns both to generate lemmas via a sequential decoder and copy the lemma characters from the external candidates supplied during run-time. Our lemmatizer enhanced with candidates extracted from the Apertium morphological analyzer achieves statistically significant improvements compared to baseline models not utilizing additional lemma information, achieves an average accuracy of 97.25% on a set of 23 UD languages, which is 0.55% higher than obtained with the Stanford Stanza model on the same set of languages. We also compare with other methods of integrating external data into lemmatization and show that our enhanced system performs considerably better than a simple lexicon extension method based on the Stanza system, and it achieves complementary improvements w.r.t. the data augmentation method.

cs.CL↗

arXiv2021

Evaluating Multilingual BERT for Estonian

Recently, large pre-trained language models, such as BERT, have reached state-of-the-art performance in many natural language processing tasks, but for many languages, including Estonian, BERT models are not yet available. However, there exist several multilingual BERT models that can handle multiple languages simultaneously and that have been trained also on Estonian data. In this paper, we evaluate four multilingual models -- multilingual BERT, multilingual distilled BERT, XLM and XLM-RoBERTa -- on several NLP tasks including POS and morphological tagging, NER and text classification. Our aim is to establish a comparison between these multilingual BERT models and the existing baseline neural models for these tasks. Our results show that multilingual BERT models can generalise well on different Estonian NLP tasks outperforming all baselines models for POS and morphological tagging and text classification, and reaching the comparable level with the best baseline for NER, with XLM-RoBERTa achieving the highest results compared with other multilingual models.

cs.CL↗

arXiv2020

Evaluating Sentence Segmentation and Word Tokenization Systems on Estonian Web Texts

Texts obtained from web are noisy and do not necessarily follow the orthographic sentence and word boundary rules. Thus, sentence segmentation and word tokenization systems that have been developed on well-formed texts might not perform so well on unedited web texts. In this paper, we first describe the manual annotation of sentence boundaries of an Estonian web dataset and then present the evaluation results of three existing sentence segmentation and word tokenization systems on this corpus: EstNLTK, Stanza and UDPipe. While EstNLTK obtains the highest performance compared to other systems on sentence segmentation on this dataset, the sentence segmentation performance of Stanza and UDPipe remains well below the results obtained on the more well-formed Estonian UD test set.

cs.CL↗

arXiv2018

Modeling Composite Labels for Neural Morphological Tagging

Neural morphological tagging has been regarded as an extension to POS tagging task, treating each morphological tag as a monolithic label and ignoring its internal structure. We propose to view morphological tags as composite labels and explicitly model their internal structure in a neural sequence tagger. For this, we explore three different neural architectures and compare their performance with both CRF and simple neural multiclass baselines. We evaluate our models on 49 languages and show that the neural architecture that models the morphological labels as sequences of morphological category values performs significantly better than both baselines establishing state-of-the-art results in morphological tagging for most languages.

cs.CL↗

arXiv2018

Neural Morphological Tagging for Estonian

We develop neural morphological tagging and disambiguation models for Estonian. First, we experiment with two neural architectures for morphological tagging - a standard multiclass classifier which treats each morphological tag as a single unit, and a sequence model which handles the morphological tags as sequences of morphological category values. Secondly, we complement these models with the analyses generated by a rule-based Estonian morphological analyser (MA) VABAMORF , thus performing a soft morphological disambiguation. We compare two ways of supplementing a neural morphological tagger with the MA outputs: firstly, by adding the combined analyses embeddings to the word representation input to the neural tagging model, and secondly, by adopting an attention mechanism to focus on the most relevant analyses generated by the MA. Experiments on three Estonian datasets show that our neural architectures consistently outperform the non-neural baselines, including HMM-disambiguated VABAMORF, while augmenting models with MA outputs results in a further performance boost for both models.

cs.CL↗

arXiv2018

The Impact of Annotation Guidelines and Annotated Data on Extracting App Features from App Reviews

Annotation guidelines used to guide the annotation of training and evaluation datasets can have a considerable impact on the quality of machine learning models. In this study, we explore the effects of annotation guidelines on the quality of app feature extraction models. As a main result, we propose several changes to the existing annotation guidelines with a goal of making the extracted app features more useful and informative to the app developers. We test the proposed changes via simulating the application of the new annotation guidelines and then evaluating the performance of the supervised machine learning models trained on datasets annotated with initial and simulated guidelines. While the overall performance of automatic app feature extraction remains the same as compared to the model trained on the dataset with initial annotations, the features extracted by the model trained on the dataset with simulated new annotations are less noisy and more informative to the app developers. Secondly, we are interested in what kind of annotated training data is necessary for training an automatic app feature extraction model. In particular, we explore whether the training set should contain annotated app reviews from those apps/app categories on which the model is subsequently planned to be applied, or is it sufficient to have annotated app reviews from any app available for training, even when these apps are from very different categories compared to the test app. Our experiments show that having annotated training reviews from the test app is not necessary although including them into training set helps to improve recall. Furthermore, we test whether augmenting the training set with annotated product reviews helps to improve the performance of app feature extraction. We find that the models trained on augmented training set lead to improved recall but at the cost of the drop in precision.

cs.IR↗

arXiv2017

Idea density for predicting Alzheimer's disease from transcribed speech

Idea Density (ID) measures the rate at which ideas or elementary predications are expressed in an utterance or in a text. Lower ID is found to be associated with an increased risk of developing Alzheimer's disease (AD) (Snowdon et al., 1996; Engelman et al., 2010). ID has been used in two different versions: propositional idea density (PID) counts the expressed ideas and can be applied to any text while semantic idea density (SID) counts pre-defined information content units and is naturally more applicable to normative domains, such as picture description tasks. In this paper, we develop DEPID, a novel dependency-based method for computing PID, and its version DEPID-R that enables to exclude repeating ideas---a feature characteristic to AD speech. We conduct the first comparison of automatically extracted PID and SID in the diagnostic classification task on two different AD datasets covering both closed-topic and free-recall domains. While SID performs better on the normative dataset, adding PID leads to a small but significant improvement (+1.7 F-score). On the free-topic dataset, PID performs better than SID as expected (77.6 vs 72.3 in F-score) but adding the features derived from the word embedding clustering underlying the automatic SID increases the results considerably, leading to an F-score of 84.8.

cs.CL↗

arXiv2017

Linear Ensembles of Word Embedding Models

This paper explores linear methods for combining several word embedding models into an ensemble. We construct the combined models using an iterative method based on either ordinary least squares regression or the solution to the orthogonal Procrustes problem. We evaluate the proposed approaches on Estonian---a morphologically complex language, for which the available corpora for training word embeddings are relatively small. We compare both combined models with each other and with the input word embedding models using synonym and analogy tests. The results show that while using the ordinary least squares regression performs poorly in our experiments, using orthogonal Procrustes to combine several word embedding models into an ensemble model leads to 7-10% relative improvements over the mean result of the initial models in synonym tests and 19-47% in analogy tests.

cs.CL↗

arXiv2017

Neighborhood Mixture Model for Knowledge Base Completion

Knowledge bases are useful resources for many natural language processing tasks, however, they are far from complete. In this paper, we define a novel entity representation as a mixture of its neighborhood in the knowledge base and apply this technique on TransE-a well-known embedding model for knowledge base completion. Experimental results show that the neighborhood information significantly helps to improve the results of the TransE model, leading to better performance than obtained by other state-of-the-art embedding models on three benchmark datasets for triple classification, entity prediction and relation prediction tasks.

cs.CL↗

arXiv2017

STransE: a novel embedding model of entities and relationships in knowledge bases

Knowledge bases of real-world facts about entities and their relationships are useful resources for a variety of natural language processing tasks. However, because knowledge bases are typically incomplete, it is useful to be able to perform link prediction or knowledge base completion, i.e., predict whether a relationship not in the knowledge base is likely to be true. This paper combines insights from several previous link prediction models into a new embedding model STransE that represents each entity as a low-dimensional vector, and each relation by two matrices and a translation vector. STransE is a simple combination of the SE and TransE models, but it obtains better link prediction performance on two benchmark datasets than previous embedding models. Thus, STransE can serve as a new baseline for the more complex models in the link prediction task.

cs.CL↗