No abstract

We train several language models for Icelandic, including IceBERT, that achieve state-of-the-art performance in a variety of downstream tasks, including part-of-speech tagging, named entity recognition, grammatical error detection and constituency parsing. To train the models we introduce a new corpus of Icelandic text, the Icelandic Common Crawl Corpus (IC3), a collection of high quality texts found online by targeting the Icelandic top-level-domain .is. Several other public data sources are also collected for a total of 16GB of Icelandic text. To enhance the evaluation of model performance and to raise the bar in baselines for Icelandic, we manually translate and adapt the WinoGrande commonsense reasoning dataset. Through these efforts we demonstrate that a properly cleaned crawled corpus is sufficient to achieve state-of-the-art results in NLP applications for low to medium resource languages, by comparison with models trained on a curated corpus. We further show that initializing models using existing multilingual models can lead to state-of-the-art results for some downstream tasks.

We present Miðeind's submission for the English→Icelandic and Icelandic→English subsets of the 2021 WMT news translation task. Transformer-base models are trained for translation on parallel data to generate backtranslations iteratively. A pretrained mBART-25 model is then adapted for translation using parallel data as well as the last backtranslation iteration. This adapted pretrained model is then used to re-generate backtranslations, and the training of the adapted model is continued.

A set of Fortran subroutines for reverse mode algorithmic (or automatic) differentiation of the basic linear algebra subprograms (BLAS) is presented. This is preceded by a description of the mathematical tools used to obtain the formulae of these derivatives, with emphasis on special matrices supported by the BLAS: triangular, symmetric, and band. All single and double precision BLAS derivatives have been implemented, together with the Cholesky factorization from Linear Algebra Package (LAPACK). The subroutines are written in Fortran 2003 with a Fortran 77 interface to allow use from C and C++, as well as dynamic languages such as R, Python, Matlab, and Octave. The subroutines are all implemented by calling BLAS, thereby attaining fast runtime. Timing results show derivative runtimes that are about twice those of the corresponding BLAS, in line with theory. The emphasis is on reverse mode because it is more important for the main application that we have in mind, numerical optimization. Two examples are presented, one dealing with the least squares modeling of groundwater, and the other dealing with the maximum likelihood estimation of the parameters of a vector autoregressive time series. The article contains comprehensive tables of formulae for the BLAS derivatives as well as for several non-BLAS matrix operations commonly used in optimization.