This paper describes a reproduction of a human evaluation study evaluating redundancies generated in automatically generated text from a data-to-text system. While the scope of the original study is broader, a human evaluation—a manual error analysis—is included as part of the system evaluation. We attempt a reproduction of this human evaluation, however while the authors annotate multiple properties of the generated text, we focus exclusively on a single quality criterion, that of redundancy. In focusing our study on a single minimal reproducible experimental unit, with the experiment being fairly straightforward and all data made available by the authors, we encountered no challenges with our reproduction and were able to reproduce the trend found in the original experiment. However, while still confirming the general trend, we found that both our annotators identified twice as many errors in the dataset than the original authors.

The goal of this paper is to learn more about how idiomatic information is structurally encoded in embeddings, using a structural probing method. We repurpose an existing English verbal multi-word expression (MWE) dataset to suit the probing framework and perform a comparative probing study of static (GloVe) and contextual (BERT) embeddings. Our experiments indicate that both encode some idiomatic information to varying degrees, but yield conflicting evidence as to whether idiomaticity is encoded in the vector norm, leaving this an open question. We also identify some limitations of the used dataset and highlight important directions for future work in improving its suitability for a probing analysis.

Modelling taxonomic and thematic relatedness is important for building AI with comprehensive natural language understanding. The goal of this paper is to learn more about how taxonomic information is structurally encoded in embeddings. To do this, we design a new hypernym-hyponym probing task and perform a comparative probing study of taxonomic and thematic SGNS and GloVe embeddings. Our experiments indicate that both types of embeddings encode some taxonomic information, but the amount, as well as the geometric properties of the encodings, are independently related to both the encoder architecture, as well as the embedding training data. Specifically, we find that only taxonomic embeddings carry taxonomic information in their norm, which is determined by the underlying distribution in the data.

This report describes a reproduction of a human evaluation study evaluating automatically detected over- and undertranslations obtained using neural machine translation approaches. While the scope of the original study is much broader, a human evaluation is included as part of its system evaluation. We attempt an exact reproduction of this human evaluation, pertaining to translations on the the English-German language pair. While encountering minor logistical challenges, with all the source material being publicly available and some additional instructions provided by the original authors, we were able to reproduce the original experiment with only minor differences in the results.

We report our efforts in identifying a set of previous human evaluations in NLP that would be suitable for a coordinated study examining what makes human evaluations in NLP more/less reproducible. We present our results and findings, which include that just 13% of papers had (i) sufficiently low barriers to reproduction, and (ii) enough obtainable information, to be considered for reproduction, and that all but one of the experiments we selected for reproduction was discovered to have flaws that made the meaningfulness of conducting a reproduction questionable. As a result, we had to change our coordinated study design from a reproduce approach to a standardise-then-reproduce-twice approach. Our overall (negative) finding that the great majority of human evaluations in NLP is not repeatable and/or not reproducible and/or too flawed to justify reproduction, paints a dire picture, but presents an opportunity for a rethink about how to design and report human evaluations in NLP.

PRINCIPLE was a Connecting Europe Facility (CEF)-funded project that focused on the identification, collection and processing of language resources (LRs) for four European under-resourced languages (Croatian, Icelandic, Irish and Norwegian) in order to improve translation quality of eTranslation, an online machine translation (MT) tool provided by the European Commission. The collected LRs were used for the development of neural MT engines in order to verify the quality of the resources. For all four languages, a total of 66 LRs were collected and made available on the ELRC-SHARE repository under various licenses. For Croatian, we have collected and published 20 LRs: 19 parallel corpora and 1 glossary. The majority of data is in the general domain (72 % of translation units), while the rest is in the eJustice (23 %), eHealth (3 %) and eProcurement (2 %) Digital Service Infrastructures (DSI) domains. The majority of the resources were for the Croatian-English language pair. The data was donated by six data contributors from the public as well as private sector. In this paper we present a subset of 13 Croatian LRs developed based on public administration documents, which are all made freely available, as well as challenges associated with the data collection, cleaning and processing.

Improving our understanding of how information is encoded in vector space can yield valuable interpretability insights. Alongside vector dimensions, we argue that it is possible for the vector norm to also carry linguistic information. We develop a method to test this: an extension of the probing framework which allows for relative intrinsic interpretations of probing results. It relies on introducing noise that ablates information encoded in embeddings, grounded in random baselines and confidence intervals. We apply the method to well-established probing tasks and find evidence that confirms the existence of separate information containers in English GloVe and BERT embeddings. Our correlation analysis aligns with the experimental findings that different encoders use the norm to encode different kinds of information: GloVe stores syntactic and sentence length information in the vector norm, while BERT uses it to encode contextual incongruity.

This is a resource description paper that describes the creation and properties of a set of pseudo-corpora generated artificially from a random walk over the English WordNet taxonomy. Our WordNet taxonomic random walk implementation allows the exploration of different random walk hyperparameters and the generation of a variety of different pseudo-corpora. We find that different combinations of parameters result in varying statistical properties of the generated pseudo-corpora. We have published a total of 81 pseudo-corpora that we have used in our previous research, but have not exhausted all possible combinations of hyperparameters, which is why we have also published a codebase that allows the generation of additional WordNet taxonomic pseudo-corpora as needed. Ultimately, such pseudo-corpora can be used to train taxonomic word embeddings, as a way of transferring taxonomic knowledge into a word embedding space.

Creating word embeddings that reflect semantic relationships encoded in lexical knowledge resources is an open challenge. One approach is to use a random walk over a knowledge graph to generate a pseudo-corpus and use this corpus to train embeddings. However, the effect of the shape of the knowledge graph on the generated pseudo-corpora, and on the resulting word embeddings, has not been studied. To explore this, we use English WordNet, constrained to the taxonomic (tree-like) portion of the graph, as a case study. We investigate the properties of the generated pseudo-corpora, and their impact on the resulting embeddings. We find that the distributions in the psuedo-corpora exhibit properties found in natural corpora, such as Zipf’s and Heaps’ law, and also observe that the proportion of rare words in a pseudo-corpus affects the performance of its embeddings on word similarity.

This paper describes a simple but competitive unsupervised system for hypernym discovery. The system uses skip-gram word embeddings with negative sampling, trained on specialised corpora. Candidate hypernyms for an input word are predicted based based on cosine similarity scores. Two sets of word embedding models were trained separately on two specialised corpora: a medical corpus and a music industry corpus. Our system scored highest in the medical domain among the competing unsupervised systems but performed poorly on the music industry domain. Our system does not depend on any external data other than raw specialised corpora.

Machine translation between closely related languages is less challenging and exibits a smaller number of translation errors than translation between distant languages, but there are still obstacles which should be addressed in order to improve such systems. This work explores the obstacles for machine translation systems between closely related South Slavic languages, namely Croatian, Serbian and Slovenian. Statistical systems for all language pairs and translation directions are trained using parallel texts from different domains, however mainly on spoken language i.e. subtitles. For translation between Serbian and Croatian, a rule-based system is also explored. It is shown that for all language pairs and translation systems, the main obstacles are differences between structural properties.

This paper presents an approach for building large monolingual corpora and, at the same time, extracting parallel data by crawling the top-level domain of a given language of interest. For gathering linguistically relevant data from top-level domains we use the SpiderLing crawler, modified to crawl data written in multiple languages. The output of this process is then fed to Bitextor, a tool for harvesting parallel data from a collection of documents. We call the system combining these two tools Spidextor, a blend of the names of its two crucial parts. We evaluate the described approach intrinsically by measuring the accuracy of the extracted bitexts from the Croatian top-level domain “.hr” and the Slovene top-level domain “.si”, and extrinsically on the English-Croatian language pair by comparing an SMT system built from the crawled data with third-party systems. We finally present parallel datasets collected with our approach for the English-Croatian, English-Finnish, English-Serbian and English-Slovene language pairs.

In this paper we present newly developed inflectional lexcions and manually annotated corpora of Croatian and Serbian. We introduce hrLex and srLex - two freely available inflectional lexicons of Croatian and Serbian - and describe the process of building these lexicons, supported by supervised machine learning techniques for lemma and paradigm prediction. Furthermore, we introduce hr500k, a manually annotated corpus of Croatian, 500 thousand tokens in size. We showcase the three newly developed resources on the task of morphosyntactic annotation of both languages by using a recently developed CRF tagger. We achieve best results yet reported on the task for both languages, beating the HunPos baseline trained on the same datasets by a wide margin.

In this paper we compare two tools for automatically harvesting bitexts from multilingual websites: bitextor and ILSP-FC. We used both tools for crawling 21 multilingual websites from the tourism domain to build a domain-specific English―Croatian parallel corpus. Different settings were tried for both tools and 10,662 unique document pairs were obtained. A sample of about 10% of them was manually examined and the success rate was computed on the collection of pairs of documents detected by each setting. We compare the performance of the settings and the amount of different corpora detected by each setting. In addition, we describe the resource obtained, both by the settings and through the human evaluation, which has been released as a high-quality parallel corpus.