The ArisToCAT project aims to assess the comprehensibility of ‘raw’ (unedited) MT output for readers who can only rely on the MT output. In this project description, we summarize the main results of the project and present future work.

Several studies (covering many language pairs and translation tasks) have demonstrated that translation quality has improved enormously since the emergence of neural machine translation systems. This raises the question whether such systems are able to produce high-quality translations for more creative text types such as literature and whether they are able to generate coherent translations on document level. Our study aimed to investigate these two questions by carrying out a document-level evaluation of the raw NMT output of an entire novel. We translated Agatha Christie’s novel The Mysterious Affair at Styles with Google’s NMT system from English into Dutch and annotated it in two steps: first all fluency errors, then all accuracy errors. We report on the overall quality, determine the remaining issues, compare the most frequent error types to those in general-domain MT, and investigate whether any accuracy and fluency errors co-occur regularly. Additionally, we assess the inter-annotator agreement on the first chapter of the novel.

In order to improve the symbiosis between machine translation (MT) system and post-editor, it is not enough to know that the output of one system is better than the output of another system. A fine-grained error analysis is needed to provide information on the type and location of errors occurring in MT and the corresponding errors occurring after post-editing (PE). This article reports on a fine-grained translation quality assessment approach which was applied to machine translated-texts and the post-edited versions of these texts, made by student post-editors. By linking each error to the corresponding source text-passage, it is possible to identify passages that were problematic in MT, but not after PE, or passages that were problematic even after PE. This method provides rich data on the origin and impact of errors, which can be used to improve post-editor training as well as machine translation systems. We present the results of a pilot experiment on the post-editing of newspaper articles and highlight the advantages of our approach.

Keystroke logging tools are a valuable aid to monitor written language production. These tools record all keystrokes, including backspaces and deletions together with timing information. In this paper we report on an extension to the keystroke logging program Inputlog in which we aggregate the logged process data from the keystroke (character) level to the word level. The logged process data are further enriched with different kinds of linguistic information: part-of-speech tags, lemmata, chunk boundaries, syllable boundaries and word frequency. A dedicated parser has been developed that distils from the logged process data word-level revisions, deleted fragments and final product data. The linguistically-annotated output will facilitate the linguistic analysis of the logged data and will provide a valuable basis for more linguistically-oriented writing process research. The set-up of the extension to Inputlog is largely language-independent. As proof-of-concept, the extension has been developed for English and Dutch. Inputlog is freely available for research purposes.

The importance of sentence-aligned parallel corpora has been widely acknowledged. Reference corpora in which sub-sentential translational correspondences are indicated manually are more labour-intensive to create, and hence less wide-spread. Such manually created reference alignments -- also called Gold Standards -- have been used in research projects to develop or test automatic word alignment systems. In most translations, translational correspondences are rather complex; for example word-by-word correspondences can be found only for a limited number of words. A reference corpus in which those complex translational correspondences are aligned manually is therefore also a useful resource for the development of translation tools and for translation studies. In this paper, we describe how we created a Gold Standard for the Dutch-English language pair. We present the annotation scheme, annotation guidelines, annotation tool and inter-annotator results. To cover a wide range of syntactic and stylistic phenomena that emerge from different writing and translation styles, our Gold Standard data set contains texts from different text types. The Gold Standard will be publicly available as part of the Dutch Parallel Corpus.

A wide spectrum of multilingual applications have aligned parallel corpora as their prerequisite. The aim of the project described in this paper is to build a multilingual corpus where all sentences are aligned at very high precision with a minimal human effort involved. The experiments on a combination of sentence aligners with different underlying algorithms described in this paper showed that by verifying only those links which were not recognized by at least two aligners, an error rate can be reduced by 93.76% as compared to the performance of the best aligner. Such manual involvement concerned only a small portion of all data (6%). This significantly reduces a load of manual work necessary to achieve nearly 100% accuracy of alignment.