We perform automatic paraphrase detection on subtitle data from the Opusparcus corpus comprising six European languages: German, English, Finnish, French, Russian, and Swedish. We train two types of supervised sentence embedding models: a word-averaging (WA) model and a gated recurrent averaging network (GRAN) model. We find out that GRAN outperforms WA and is more robust to noisy training data. Better results are obtained with more and noisier data than less and cleaner data. Additionally, we experiment on other datasets, without reaching the same level of performance, because of domain mismatch between training and test data.

We present new state-of-the-art benchmarks for paraphrase detection on all six languages in the Opusparcus sentential paraphrase corpus: English, Finnish, French, German, Russian, and Swedish. We reach these baselines by finetuning BERT. The best results are achieved on smaller and cleaner subsets of the training sets than was observed in previous research. Additionally, we study a translation-based approach that is competitive for the languages with more limited and noisier training data.

No abstract

We perform neural machine translation of sentence fragments in order to create large amounts of training data for English grammatical error correction. Our method aims at simulating mistakes made by second language learners, and produces a wider range of non-native style language in comparison to state-of-the-art synthetic data creation methods. In addition to purely grammatical errors, our approach generates other types of errors, such as lexical errors. We perform grammatical error correction experiments using neural sequence-to-sequence models, and carry out quantitative and qualitative evaluation. A model trained on data created using our proposed method is shown to outperform a baseline model on test data with a high proportion of errors.

Electrophysiological studies show that top-down modulation enhances neural tracking of attended speech in environments with overlapping speech. Yet, the specific cortical regions involved remain unclear due to the limited spatial resolution of most electrophysiological techniques. Therefore, we performed speech envelope reconstruction and representational dissimilarity-based EEG-fMRI fusion (using temporal response function estimated from EEG, n = 19, and fMRI, n = 19) to determine the spatiotemporal dynamics of attention to audiovisual cocktail-party speech. Attention related enhancement of neural tracking fluctuated in predictable temporal profiles. Such temporal dynamics may arise due to interactions between attention and prediction or other plastic mechanisms in the auditory cortex, or both. EEG-fMRI fusion revealed attention-related recurrent feedforward-feedback loops in the ventral processing stream. Our findings support models where attention facilitates dynamic neural changes in the auditory cortex, ultimately aiding discrimination of relevant sounds from irrelevant ones using minimal neural resources.