Due to the wide-spread development of Machine Translation (MT) systems –especially Neural Machine Translation (NMT) systems– MT evaluation, both automatic and human, has become more and more important as it helps us establish how MT systems perform. Yet, automatic evaluation metrics have lagged behind, as the most popular choices (e.g., BLEU, METEOR and ROUGE) may correlate poorly with human judgments. This paper seeks to put to the test an evaluation model based on a novel deep learning schema (NoDeeLe) used to compare two NMT systems on four different text genres, i.e. medical, legal, marketing and literary in the English-Greek language pair. The model utilizes information from the source segments, the MT outputs and the reference translation, as well as the automatic metrics BLEU, METEOR and WER. The proposed schema achieves a strong correlation with human judgment (78% average accuracy for the four texts with the highest accuracy, i.e. 85%, observed in the case of the marketing text), while it outperforms classic machine learning algorithms and automatic metrics.

The present study aims to compare three systems: a generic statistical machine translation (SMT), a generic neural machine translation (NMT) and a tailored-NMT system focusing on the English to Greek language pair. The comparison is carried out following a mixed-methods approach, i.e. automatic metrics, as well as side-by-side ranking, adequacy and fluency rating, measurement of actual post editing (PE) effort and human error analysis performed by 16 postgraduate Translation students. The findings reveal a higher score for both the generic NMT and the tailored-NMT outputs as regards automatic metrics and human evaluation metrics, with the tailored-NMT output faring even better than the generic NMT output.

The increasing volume of communication via microblogging messages on social networks has created the need for efficient Natural Language Processing (NLP) tools, especially for unstructured text processing. Extracting information from unstructured social text is one of the most demanding NLP tasks. This paper presents the first part-of-speech tagged data set of social text in Greek, as well as the first supervised part-of-speech tagger developed for such data sets.

The automatic evaluation of machine translation (MT) has proven to be a very significant research topic. Most automatic evaluation methods focus on the evaluation of the output of MT as they compute similarity scores that represent translation quality. This work targets on the performance of MT evaluation. We present a general scheme for learning to classify parallel translations, using linguistic information, of two MT model outputs and one human (reference) translation. We present three experiments to this scheme using neural networks (NN). One using string based hand-crafted features (Exp1), the second using automatically trained embeddings from the reference and the two MT outputs (one from a statistical machine translation (SMT) model and the other from a neural ma-chine translation (NMT) model), which are learned using NN (Exp2), and the third experiment (Exp3) that combines information from the other two experiments. The languages involved are English (EN), Greek (GR) and Italian (IT) segments are educational in domain. The proposed language-independent learning scheme which combines information from the two experiments (experiment 3) achieves higher classification accuracy compared with models using BLEU score information as well as other classification approaches, such as Random Forest (RF) and Support Vector Machine (SVM).

The present work is an overview of the TraMOOC (Translation for Massive Open Online Courses) research and innovation project, a machine translation approach for online educational content. More specifically, videolectures, assignments, and MOOC forum text is automatically translated from English into eleven European and BRIC languages. Unlike previous approaches to machine translation, the output quality in TraMOOC relies on a multimodal evaluation schema that involves crowdsourcing, error type markup, an error taxonomy for translation model comparison, and implicit evaluation via text mining, i.e. entity recognition and its performance comparison between the source and the translated text, and sentiment analysis on the students’ forum posts. Finally, the evaluation output will result in more and better quality in-domain parallel data that will be fed back to the translation engine for higher quality output. The translation service will be incorporated into the Iversity MOOC platform and into the VideoLectures.net digital library portal.

This paper describes Eksairesis, a system for learning economic domain knowledge automatically from Modern Greek text. The knowledge is in the form of economic terms and the semantic relations that govern them. The entire process in based on the use of minimal language-dependent tools, no external linguistic resources, and merely free, unstructured text. The methodology is thereby easily portable to other domains and other languages. The text is pre-processed with basic morphological annotation, and semantic (named and other) entities are identified using supervised learning techniques. Statistical filtering, i.e. corpora comparison is used to extract domain terms and supervised learning is again employed to detect the semantic relations between pairs of terms. Advanced classification schemata, ensemble learning, and one-sided sampling, are experimented with in order to deal with the noise in the data, which is unavoidable due to the low pre-processing level and the lack of sophisticated resources. An average 68.5% f-score over all the classes is achieved when learning semantic relations. Bearing in mind the use of minimal resources and the highly automated nature of the process, classification performance is very promising, compared to results reported in previous work.

For the present work, we deal with the significant problem of high imbalance in data in binary or multi-class classification problems. We study two different linguistic applications. The former determines whether a syntactic construction (environment) co-occurs with a verb in a natural text corpus consists a subcategorization frame of the verb or not. The latter is called Name Entity Recognition (NER) and it concerns determining whether a noun belongs to a specific Name Entity class. Regarding the subcategorization domain, each environment is encoded as a vector of heterogeneous attributes, where a very high imbalance between positive and negative examples is observed (an imbalance ratio of approximately 1:80). In the NER application, the imbalance between a name entity class and the negative class is even greater (1:120). In order to confront the plethora of negative instances, we suggest a search tactic during training phase that employs Tomek links for reducing unnecessary negative examples from the training set. Regarding the classification mechanism, we argue that Bayesian networks are well suited and we propose a novel network structure which efficiently handles heterogeneous attributes without discretization and is more classification-oriented. Comparing the experimental results with those of other known machine learning algorithms, our methodology performs significantly better in detecting examples of the rare class.