The performance of NLP methods for severely under-resourced languages cannot currently hope to match the state of the art in NLP methods for well resourced languages. We explore the extent to which pretrained large language models (LLMs) can bridge this gap, via the example of data-to-text generation for Irish, Welsh, Breton and Maltese. We test LLMs on these under-resourced languages and English, in a range of scenarios. We find that LLMs easily set the state of the art for the under-resourced languages by substantial margins, as measured by both automatic and human evaluations. For all our languages, human evaluation shows on-a-par performance with humans for our best systems, but BLEU scores collapse compared to English, casting doubt on the metric’s suitability for evaluating non-task-specific systems. Overall, our results demonstrate the great potential of LLMs to bridge the performance gap for under-resourced languages.

As the cost of training ever larger language models has grown, so has the interest in reusing previously learnt knowledge. Transfer learning methods have shown how reusing non-task-specific knowledge can help in subsequent task-specific learning.In this paper, we investigate the inverse: porting whole functional modules that encode task-specific knowledge from one model to another. We designed a study comprising 1,440 training/testing runs to test the portability of modules trained by parameter-efficient finetuning (PEFT) techniques, using sentiment analysis as an example task. We test portability in a wide range of scenarios, involving different PEFT techniques and different pretrained host models, among other dimensions. We compare the performance of ported modules with that of equivalent modules trained (i) from scratch, and (ii) from parameters sampled from the same distribution as the ported module.We find that the ported modules far outperform the two alternatives tested, but that there are interesting differences between the four PEFT techniques tested.We conclude that task-specific knowledge in the form of structurally modular sets of parameters as produced by PEFT techniques is highly portable, but that degree of success depends on type of PEFT and on differences between originating and receiving pretrained models.

Human evaluation is widely considered the most reliable form of evaluation in NLP, but recent research has shown it to be riddled with mistakes, often as a result of manual execution of tasks. This paper argues that such mistakes could be avoided if we were to automate, as much as is practical, the process of performing experiments for human evaluation of NLP systems. We provide a simple methodology that can improve both the transparency and reproducibility of experiments. We show how the sequence of component processes of a human evaluation can be defined in advance, facilitating full or partial automation, detailed preregistration of the process, and research transparency and repeatability.

In this paper, we investigate how different semantic, or content-related, errors made by different types of data-to-text systems differ in terms of number and type. In total, we examine 15 systems: three rule-based and 12 neural systems including two large language models without training or fine-tuning. All systems were tested on the English WebNLG dataset version 3.0. We use a semantic error taxonomy and the brat annotation tool to obtain word-span error annotations on a sample of system outputs. The annotations enable us to establish how many semantic errors different (types of) systems make and what specific types of errors they make, and thus to get an overall understanding of semantic strengths and weaknesses among various types of NLG systems. Among our main findings, we observe that symbolic (rule and template-based) systems make fewer semantic errors overall, non-LLM neural systems have better fluency and data coverage, but make more semantic errors, while LLM-based systems require improvement particularly in addressing superfluous.

Four years on from two papers (Belz et al., 2020; Howcroft et al., 2020) that first called out the lack of standardisation and comparability in the quality criteria assessed in NLP system evaluations, researchers still use widely differing quality criteria names and definitions, meaning that it continues to be unclear when the same aspect of quality is being assessed in two evaluations. While normalised quality criteria were proposed at the time, the list was unwieldy and using it came with a steep learning curve. In this demo paper, our aim is to address these issues with an interactive taxonomy tool that enables quick perusal and selection of the quality criteria, and provides decision support and examples of use at each node.

Wikipedia is known to have systematic gaps in its coverage that correspond to under-resourced languages as well as underrepresented groups. This paper presents a new tool to support efforts to fill in these gaps by automatically generating draft articles and facilitating post-editing and uploading to Wikipedia. A rule-based generator and an input-constrained LLM are used to generate two alternative articles, enabling the often more fluent, but error-prone, LLM-generated article to be content-checked against the more reliable, but less fluent, rule-generated article.

Following numerous calls in the literature for improved practices and standardisation in human evaluation in Natural Language Processing over the past ten years, we held a tutorial on the topic at the 2024 INLG Conference. The tutorial addressed the structure, development, design, implementation, execution and analysis of human evaluations of NLP system quality. Hands-on practical sessions were run, designed to facilitate assimilation of the material presented. Slides, lecture recordings, code and data have been made available on GitHub (https://github.com/Human-Evaluation-Tutorial/INLG-2024-Tutorial). In this paper, we provide summaries of the content of the eight units of the tutorial, alongside its research context and aims.

While conducting a coordinated set of repeat runs of human evaluation experiments in NLP, we discovered flaws in every single experiment we selected for inclusion via a systematic process. In this squib, we describe the types of flaws we discovered, which include coding errors (e.g., loading the wrong system outputs to evaluate), failure to follow standard scientific practice (e.g., ad hoc exclusion of participants and responses), and mistakes in reported numerical results (e.g., reported numbers not matching experimental data). If these problems are widespread, it would have worrying implications for the rigor of NLP evaluation experiments as currently conducted. We discuss what researchers can do to reduce the occurrence of such flaws, including pre-registration, better code development practices, increased testing and piloting, and post-publication addressing of errors.

This paper presents an overview of, and the results from, the 2024 Shared Task on Reproducibility of Evaluations in NLP (ReproNLP’24), following on from three previous shared tasks on reproducibility of evaluations in NLP, ReproNLP’23, ReproGen’22 and ReproGen’21. This shared task series forms part of an ongoing research programme designed to develop theory and practice of reproducibility assessment in NLP and machine learning, against a backdrop of increasing recognition of the importance of reproducibility across the two fields. We describe the ReproNLP’24 shared task, summarise results from the reproduction studies submitted, and provide additional comparative analysis of their results.

Rerunning a metric-based evaluation should be more straightforward and results should be closer than in a human-based evaluation, especially where code and model checkpoints are made available by the original authors. As this brief report of our efforts to rerun a metric-based evaluation of a set of multi-aspect controllable text generation (CTG) techniques shows however, such reruns of evaluations do not always produce results that are the same as the original results, and can reveal errors in the orginal work.

The biomedical field relies on cost and time intensive systematic reviews of papers to enable practitioners to keep up to date with research. Impressive recent advances in large language models (LLMs) have made the task of automating at least part of the systematic review process feasible, but progress is slow. This paper identifies some factors that may have been holding research back, and proposes a new, enhanced dataset and prompting-based method for automatic synthesis generation, the most challenging step for automation. We test different models and types of information from and about biomedical studies for their usefulness in obtaining high-quality results.We find that, surprisingly, inclusion of paper abstracts can worsens results. Instead, study summary information, and system instructions informed by domain knowledge, are key to producing high-quality syntheses.

It might reasonably be expected that running multiple experiments for the same task using the same data and model would yield very similar results. Recent research has, however, shown this not to be the case for many NLP experiments. In this paper, we report extensive coordinated work by two NLP groups to run the training and testing pipeline for three neural text simplification models under varying experimental conditions, including different random seeds, run-time environments, and dependency versions, yielding a large number of results for each of the three models using the same data and train/dev/test set splits. From one perspective, these results can be interpreted as shedding light on the reproducibility of evaluation results for the three NTS models, and we present an in-depth analysis of the variation observed for different combinations of experimental conditions. From another perspective, the results raise the question of whether the averaged score should be considered the ‘true’ result for each model.

Human evaluation is widely regarded as the litmus test of quality in NLP. A basic requirementof all evaluations, but in particular where they are used for meta-evaluation, is that they should support the same conclusions if repeated. However, the reproducibility of human evaluations is virtually never queried, let alone formally tested, in NLP which means that their repeatability and the reproducibility of their results is currently an open question. This focused contribution reports our review of human evaluation experiments reported in NLP papers over the past five years which we assessed in terms oftheir ability to be rerun. Overall, we estimatethat just 5% of human evaluations are repeatable in the sense that (i) there are no prohibitivebarriers to repetition, and (ii) sufficient information about experimental design is publicly available for rerunning them. Our estimate goesup to about 20% when author help is sought. We complement this investigation with a survey of results concerning the reproducibilityof human evaluations where those are repeatable in the first place. Here we find worryinglylow degrees of reproducibility, both in terms ofsimilarity of scores and of findings supportedby them. We summarise what insights can begleaned so far regarding how to make humanevaluations in NLP more repeatable and morereproducible.

This paper presents an overview of, and the results from, the 2023 Shared Task on Reproducibility of Evaluations in NLP (ReproNLP’23), following on from two previous shared tasks on reproducibility of evaluations in NLG, ReproGen’21 and ReproGen’22. This shared task series forms part of an ongoing research programme designed to develop theory and practice of reproducibility assessment in NLP and machine learning, all against a background of an interest in reproducibility that con- tinues to grow in the two fields. This paper describes the ReproNLP’23 shared task, summarises results from the reproduction studies submitted, and provides comparative analysis of the results.

Recent advances in the development of large Pretrained Language Models, such as GPT, BERT and Bloom, have achieved remarkable performance on a wide range of different NLP tasks. However, when used for text generation tasks, these models still have limitations when it comes to controlling the content and style of the generated text, often producing content that is incorrect, irrelevant, or inappropriate in the context of a given task. In this survey paper, we explore methods for controllable text generation with a focus on sentiment control. We systematically collect papers from the ACL Anthology, create a categorisation scheme based on different control techniques and controlled attributes, and use the scheme to categorise and compare methods. The result is a detailed and comprehensive overview of state-of-the-art techniques for sentiment-controlled text generation categorised on the basis of how the control is implemented and what attributes are controlled and providing a clear idea of their relative strengths and weaknesses.

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.

In this paper, we describe M-FleNS, a multilingual flexible plug-and-play architecture designed to accommodate neural and symbolic modules, and initially instantiated with rule-based modules. We focus on using M-FleNS for the specific purpose of building new resources for Irish, a language currently under-represented in the NLP landscape. We present the general M-FleNS framework and how we use it to build an Irish Natural Language Generation system for verbalising part of the DBpedia ontology and building a multilayered dataset with rich linguistic annotations. Via automatic and human assessments of the output texts we show that with very limited resources we are able to create a system that reaches high levels of fluency and semantic accuracy, while having very low energy and memory requirements.

Rule-based text generators lack the coverage and fluency of their neural counterparts, but have two big advantages over them: (i) they are entirely controllable and do not hallucinate; and (ii) they can fully explain how an output was generated from an input. In this paper we leverage these two advantages to create large and reliable synthetic datasets with multiple human-intelligible intermediate representations. We present the Modular Data-to-Text (Mod-D2T) Dataset which incorporates ten intermediate-level representations between input triple sets and output text; the mappings from one level to the next can broadly be interpreted as the traditional modular tasks of an NLG pipeline. We describe the Mod-D2T dataset, evaluate its quality via manual validation and discuss its applications and limitations. Data, code and documentation are available at https://github.com/mille-s/Mod-D2T.

Error analysis aims to provide insights into system errors at different levels of granularity. NLP as a field has a long-standing tradition of analysing and reporting errors which is generally considered good practice. There are existing error taxonomies tailored for different types of NLP task. In this paper, we report our work reviewing existing research on meaning/content error types in generated text, attempt to identify emerging consensus among existing meaning/content error taxonomies, and propose a standardised error taxonomy on this basis. We find that there is virtually complete agreement at the highest taxonomic level where errors of meaning/content divide into (1) Content Omission, (2) Content Addition, and (3) Content Substitution. Consensus in the lower levels is less pronounced, but a compact standardised consensus taxonomy can nevertheless be derived that works across generation tasks and application domains.

We present work in progress that aims to address the coverage issue faced by rule-based text generators. We propose a pipeline for extracting abstract dependency template (predicate-argument structures) from Wikipedia text to be used as input for generating text from structured data with the FORGe system. The pipeline comprises three main components: (i) candidate sentence retrieval, (ii) clause extraction, ranking and selection, and (iii) conversion to predicate-argument form. We present an approach and preliminary evaluation for the ranking and selection module.

The WebNLG task consists of mapping a knowledge graph to a text verbalising the con- tent of that graph. The 2017 WebNLG edi- tion required participating systems to gener- ate English text from a set of DBpedia triples, while the 2020 WebNLG+ challenge addition- ally included generation into Russian and se- mantic parsing of English and Russian texts. In contrast, WebNLG 2023 focuses on four under-resourced languages which are severely under-represented in research on text genera- tion, namely Breton, Irish, Maltese and Welsh. In addition, WebNLG 2023 once again includes Russian. In this paper, we present the organi- sation of the shared task (data, timeline, eval- uation), briefly describe the participating sys- tems and summarise results for participating systems.

LLMs are great at tasks involving English which dominates in their training data. We explore their ability to address tasks involving languages that are severely under-represented in their training data. More specifically, we do this in the context of data-to-text generation for Irish, Maltese, Welsh and Breton. During the prompt-engineering phase we tested GPT-3.5 and~4 with a range of prompt types and formats on a small sample of example input/output pairs. We then fully evaluated the two most promising prompts in two scenarios: (i) direct generation into the under-resourced languages, and (ii) generation into English followed by translation into the under-resourced languages. We find that few-shot prompting works better for direct generation into under-resourced languages, but that the difference disappears when pivoting via English. The few-shot + translation system variants were submitted to the WebNLG 2023 shared task where they outperformed all other systems by substantial margins in all languages on all automatic metrics. We conclude that good performance can be achieved with state-of-the-art LLMs out-of-the box for under-resourced languages. However, best results (for Welsh) of BLEU 25.12, ChrF++ 0.55, and TER 0.64 are well below the lowest ranked English system at WebNLG’20 with BLEU 0.391, ChrF++ 0.579, and TER 0.564.

In this paper, we describe the submission of Dublin City University (DCU) and Trinity College Dublin (TCD) for the WebNLG 2023 shared task. We present a fully rule-based pipeline for generating Irish texts from DBpedia triple sets which comprises 4 components: triple lexicalisation, generation of noninflected Irish text, inflection generation, and post-processing.

Against a background of growing interest in reproducibility in NLP and ML, and as part of an ongoing research programme designed to develop theory and practice of reproducibility assessment in NLP, we organised the second shared task on reproducibility of evaluations in NLG, ReproGen 2022. This paper describes the shared task, summarises results from the reproduction studies submitted, and provides further comparative analysis of the results. Out of six initial team registrations, we received submissions from five teams. Meta-analysis of the five reproduction studies revealed varying degrees of reproducibility, and allowed further tentative conclusions about what types of evaluation tend to have better reproducibility.

In this paper, we present the results of two reproduction studies for the human evaluation originally reported by Dušek and Kasner (2020) in which the authors comparatively evaluated outputs produced by a semantic error detection system for data-to-text generation against reference outputs. In the first reproduction, the original evaluators repeat the evaluation, in a test of the repeatability of the original evaluation. In the second study, two new evaluators carry out the evaluation task, in a test of the reproducibility of the original evaluation under otherwise identical conditions. We describe our approach to reproduction, and present and analyse results, finding different degrees of reproducibility depending on result type, data and labelling task. Our resources are available and open-sourced.

In this paper we describe our reproduction study of the human evaluation of text simplic- ity reported by Nisioi et al. (2017). The work was carried out as part of the ReproGen Shared Task 2022 on Reproducibility of Evaluations in NLG. Our aim was to repeat the evaluation of simplicity for nine automatic text simplification systems with a different set of evaluators. We describe our experimental design together with the known aspects of the original experimental design and present the results from both studies. Pearson correlation between the original and reproduction scores is moderate to high (0.776). Inter-annotator agreement in the reproduction study is lower (0.40) than in the original study (0.66). We discuss challenges arising from the unavailability of certain aspects of the origi- nal set-up, and make several suggestions as to how reproduction of similar evaluations can be made easier in future.

This work examines different ways of aggregating scores for error annotation in MT outputs: raw error counts, error counts normalised over total number of words (word percentage’), and error counts normalised over total number of errors (error percentage’). We use each of these three scores to calculate inter-annotator agreement in the form of Krippendorff’s alpha and Pearson’s r and compare the obtained numbers, overall and separately for different types of errors. While each score has its advantages depending on the goal of the evaluation, we argue that the best way of estimating inter-annotator agreement using such numbers are raw counts. If the annotation process ensures that the total number of words cannot differ among the annotators (for example, due to adding omission symbols), normalising over number of words will lead to the same conclusions. In contrast, total number of errors is very subjective because different annotators often perceive different amount of errors in the same text, therefore normalising over this number can indicate lower agreements.

While automatically computing numerical scores remains the dominant paradigm in NLP system evaluation, error analysis is receiving increasing attention, with numerous error annotation schemes being proposed for automatically generated text. However, there is little agreement about what error annotation schemes should look like, how many different types of errors should be distinguished and at what level of granularity. In this paper, our aim is to map out recent work on annotating errors in automatically generated text, with a particular focus on error taxonomies. We describe our systematic paper selection process, and survey the error annotation schemes reported in the papers, drawing out similarities and differences between them. Finally, we characterise the issues that would make it difficult to move from the current situation to a standardised error taxonomy for annotating errors in automatically generated text.

A growing body of work uses Natural Language Processing (NLP) methods to automatically generate medical notes from audio recordings of doctor-patient consultations. However, there are very few studies on how such systems could be used in clinical practice, how clinicians would adjust to using them, or how system design should be influenced by such considerations. In this paper, we present three rounds of user studies, carried out in the context of developing a medical note generation system. We present, analyse and discuss the participating clinicians’ impressions and views of how the system ought to be adapted to be of value to them. Next, we describe a three-week test run of the system in a live telehealth clinical practice. Major findings include (i) the emergence of five different note-taking behaviours; (ii) the importance of the system generating notes in real time during the consultation; and (iii) the identification of a number of clinical use cases that could prove challenging for automatic note generation systems.

This paper describes and tests a method for carrying out quantified reproducibility assessment (QRA) that is based on concepts and definitions from metrology. QRA produces a single score estimating the degree of reproducibility of a given system and evaluation measure, on the basis of the scores from, and differences between, different reproductions. We test QRA on 18 different system and evaluation measure combinations (involving diverse NLP tasks and types of evaluation), for each of which we have the original results and one to seven reproduction results. The proposed QRA method produces degree-of-reproducibility scores that are comparable across multiple reproductions not only of the same, but also of different, original studies. We find that the proposed method facilitates insights into causes of variation between reproductions, and as a result, allows conclusions to be drawn about what aspects of system and/or evaluation design need to be changed in order to improve reproducibility.

In recent years, machine learning models have rapidly become better at generating clinical consultation notes; yet, there is little work on how to properly evaluate the generated consultation notes to understand the impact they may have on both the clinician using them and the patient’s clinical safety. To address this we present an extensive human evaluation study of consultation notes where 5 clinicians (i) listen to 57 mock consultations, (ii) write their own notes, (iii) post-edit a number of automatically generated notes, and (iv) extract all the errors, both quantitative and qualitative. We then carry out a correlation study with 18 automatic quality metrics and the human judgements. We find that a simple, character-based Levenshtein distance metric performs on par if not better than common model-based metrics like BertScore. All our findings and annotations are open-sourced.

This paper presents the Human Evaluation Datasheet (HEDS), a template for recording the details of individual human evaluation experiments in Natural Language Processing (NLP), and reports on first experience of researchers using HEDS sheets in practice. Originally taking inspiration from seminal papers by Bender and Friedman (2018), Mitchell et al. (2019), and Gebru et al. (2020), HEDS facilitates the recording of properties of human evaluations in sufficient detail, and with sufficient standardisation, to support comparability, meta-evaluation,and reproducibility assessments for human evaluations. These are crucial for scientifically principled evaluation, but the overhead of completing a detailed datasheet is substantial, and we discuss possible ways of addressing this and other issues observed in practice.

Evaluating automatically generated text is generally hard due to the inherently subjective nature of many aspects of the output quality. This difficulty is compounded in automatic consultation note generation by differing opinions between medical experts both about which patient statements should be included in generated notes and about their respective importance in arriving at a diagnosis. Previous real-world evaluations of note-generation systems saw substantial disagreement between expert evaluators. In this paper we propose a protocol that aims to increase objectivity by grounding evaluations in Consultation Checklists, which are created in a preliminary step and then used as a common point of reference during quality assessment. We observed good levels of inter-annotator agreement in a first evaluation study using the protocol; further, using Consultation Checklists produced in the study as reference for automatic metrics such as ROUGE or BERTScore improves their correlation with human judgements compared to using the original human note.

Reproducibility has become an increasingly debated topic in NLP and ML over recent years, but so far, no commonly accepted definitions of even basic terms or concepts have emerged. The range of different definitions proposed within NLP/ML not only do not agree with each other, they are also not aligned with standard scientific definitions. This article examines the standard definitions of repeatability and reproducibility provided by the meta-science of metrology, and explores what they imply in terms of how to assess reproducibility, and what adopting them would mean for reproducibility assessment in NLP/ML. It turns out the standard definitions lead directly to a method for assessing reproducibility in quantified terms that renders results from reproduction studies comparable across multiple reproductions of the same original study, as well as reproductions of different original studies. The article considers where this method sits in relation to other aspects of NLP work one might wish to assess in the context of reproducibility.

Against the background of what has been termed a reproducibility crisis in science, the NLP field is becoming increasingly interested in, and conscientious about, the reproducibility of its results. The past few years have seen an impressive range of new initiatives, events and active research in the area. However, the field is far from reaching a consensus about how reproducibility should be defined, measured and addressed, with diversity of views currently increasing rather than converging. With this focused contribution, we aim to provide a wide-angle, and as near as possible complete, snapshot of current work on reproducibility in NLP,

The NLP field has recently seen a substantial increase in work related to reproducibility of results, and more generally in recognition of the importance of having shared definitions and practices relating to evaluation. Much of the work on reproducibility has so far focused on metric scores, with reproducibility of human evaluation results receiving far less attention. As part of a research programme designed to develop theory and practice of reproducibility assessment in NLP, we organised the first shared task on reproducibility of human evaluations, ReproGen 2021. This paper describes the shared task in detail, summarises results from each of the reproduction studies submitted, and provides further comparative analysis of the results. Out of nine initial team registrations, we received submissions from four teams. Meta-analysis of the four reproduction studies revealed varying degrees of reproducibility, and allowed very tentative first conclusions about what types of evaluation tend to have better reproducibility.

This paper reports results from a reproduction study in which we repeated the human evaluation of the PASS Dutch-language football report generation system (van der Lee et al., 2017). The work was carried out as part of the ReproGen Shared Task on Reproducibility of Human Evaluations in NLG, in Track A (Paper 1). We aimed to repeat the original study exactly, with the main difference that a different set of evaluators was used. We describe the study design, present the results from the original and the reproduction study, and then compare and analyse the differences between the two sets of results. For the two ‘headline’ results of average Fluency and Clarity, we find that in both studies, the system was rated more highly for Clarity than for Fluency, and Clarity had higher standard deviation. Clarity and Fluency ratings were higher, and their standard deviations lower, in the reproduction study than in the original study by substantial margins. Clarity had a higher degree of reproducibility than Fluency, as measured by the coefficient of variation. Data and code are publicly available.

In this paper we report our reproduction study of the Croatian part of an annotation-based human evaluation of machine-translated user reviews (Popovic, 2020). The work was carried out as part of the ReproGen Shared Task on Reproducibility of Human Evaluation in NLG. Our aim was to repeat the original study exactly, except for using a different set of evaluators. We describe the experimental design, characterise differences between original and reproduction study, and present the results from each study, along with analysis of the similarity between them. For the six main evaluation results of Major/Minor/All Comprehension error rates and Major/Minor/All Adequacy error rates, we find that (i) 4/6 system rankings are the same in both studies, (ii) the relative differences between systems are replicated well for Major Comprehension and Adequacy (Pearson’s > 0.9), but not for the corresponding Minor error rates (Pearson’s 0.36 for Adequacy, 0.67 for Comprehension), and (iii) the individual system scores for both types of Minor error rates had a higher degree of reproducibility than the corresponding Major error rates. We also examine inter-annotator agreement and compare the annotations obtained in the original and reproduction studies.

Human assessment remains the most trusted form of evaluation in NLG, but highly diverse approaches and a proliferation of different quality criteria used by researchers make it difficult to compare results and draw conclusions across papers, with adverse implications for meta-evaluation and reproducibility. In this paper, we present (i) our dataset of 165 NLG papers with human evaluations, (ii) the annotation scheme we developed to label the papers for different aspects of evaluations, (iii) quantitative analyses of the annotations, and (iv) a set of recommendations for improving standards in evaluation reporting. We use the annotations as a basis for examining information included in evaluation reports, and levels of consistency in approaches, experimental design and terminology, focusing in particular on the 200+ different terms that have been used for evaluated aspects of quality. We conclude that due to a pervasive lack of clarity in reports and extreme diversity in approaches, human evaluation in NLG presents as extremely confused in 2020, and that the field is in urgent need of standard methods and terminology.

Current standards for designing and reporting human evaluations in NLP mean it is generally unclear which evaluations are comparable and can be expected to yield similar results when applied to the same system outputs. This has serious implications for reproducibility testing and meta-evaluation, in particular given that human evaluation is considered the gold standard against which the trustworthiness of automatic metrics is gauged. %and merging others, as well as deciding which evaluations should be able to reproduce each other’s results. Using examples from NLG, we propose a classification system for evaluations based on disentangling (i) what is being evaluated (which aspect of quality), and (ii) how it is evaluated in specific (a) evaluation modes and (b) experimental designs. We show that this approach provides a basis for determining comparability, hence for comparison of evaluations across papers, meta-evaluation experiments, reproducibility testing.

Across NLP, a growing body of work is looking at the issue of reproducibility. However, replicability of human evaluation experiments and reproducibility of their results is currently under-addressed, and this is of particular concern for NLG where human evaluations are the norm. This paper outlines our ideas for a shared task on reproducibility of human evaluations in NLG which aims (i) to shed light on the extent to which past NLG evaluations are replicable and reproducible, and (ii) to draw conclusions regarding how evaluations can be designed and reported to increase replicability and reproducibility. If the task is run over several years, we hope to be able to document an overall increase in levels of replicability and reproducibility over time.

This paper presents results from the Third Shared Task on Multilingual Surface Realisation (SR’20) which was organised as part of the COLING’20 Workshop on Multilingual Surface Realisation. As in SR’18 and SR’19, the shared task comprised two tracks: (1) a Shallow Track where the inputs were full UD structures with word order information removed and tokens lemmatised; and (2) a Deep Track where additionally, functional words and morphological information were removed. Moreover, each track had two subtracks: (a) restricted-resource, where only the data provided or approved as part of a track could be used for training models, and (b) open-resource, where any data could be used. The Shallow Track was offered in 11 languages, whereas the Deep Track in 3 ones. Systems were evaluated using both automatic metrics and direct assessment by human evaluators in terms of Readability and Meaning Similarity to reference outputs. We present the evaluation results, along with descriptions of the SR’19 tracks, data and evaluation methods, as well as brief summaries of the participating systems. For full descriptions of the participating systems, please see the separate system reports elsewhere in this volume.

Approaches to knowledge extraction (KE) in the health domain often start by annotating text to indicate the knowledge to be extracted, and then use the annotated text to train systems to perform the KE. This may work for annotat- ing named entities or other contiguous noun phrases (drugs, some drug effects), but be- comes increasingly difficult when items tend to be expressed across multiple, possibly non- contiguous, syntactic constituents (e.g. most descriptions of drug effects in user-generated text). Other issues include that it is not al- ways clear how annotations map to actionable insights, or how they scale up to, or can form part of, more complex KE tasks. This paper reports our efforts in developing an approach to extracting knowledge about drug nonadher- ence from health forums which led us to con- clude that development cannot proceed in sep- arate steps but that all aspects—from concep- tualisation to annotation scheme development, annotation, KE system training and knowl- edge graph instantiation—are interdependent and need to be co-developed. Our aim in this paper is two-fold: we describe a generally ap- plicable framework for developing a KE ap- proach, and present a specific KE approach, developed with the framework, for the task of gathering information about antidepressant drug nonadherence. We report the conceptual- isation, the annotation scheme, the annotated corpus, and an analysis of annotated texts.

We report results from the SR’19 Shared Task, the second edition of a multilingual surface realisation task organised as part of the EMNLP’19 Workshop on Multilingual Surface Realisation. As in SR’18, the shared task comprised two tracks with different levels of complexity: (a) a shallow track where the inputs were full UD structures with word order information removed and tokens lemmatised; and (b) a deep track where additionally, functional words and morphological information were removed. The shallow track was offered in eleven, and the deep track in three languages. Systems were evaluated (a) automatically, using a range of intrinsic metrics, and (b) by human judges in terms of readability and meaning similarity. This report presents the evaluation results, along with descriptions of the SR’19 tracks, data and evaluation methods. For full descriptions of the participating systems, please see the separate system reports elsewhere in this volume.

We report results from the SR’18 Shared Task, a new multilingual surface realisation task organised as part of the ACL’18 Workshop on Multilingual Surface Realisation. As in its English-only predecessor task SR’11, the shared task comprised two tracks with different levels of complexity: (a) a shallow track where the inputs were full UD structures with word order information removed and tokens lemmatised; and (b) a deep track where additionally, functional words and morphological information were removed. The shallow track was offered in ten, and the deep track in three languages. Systems were evaluated (a) automatically, using a range of intrinsic metrics, and (b) by human judges in terms of readability and meaning similarity. This report presents the evaluation results, along with descriptions of the SR’18 tracks, data and evaluation methods. For full descriptions of the participating systems, please see the separate system reports elsewhere in this volume.

We present SpatialVOC2K, the first multilingual image dataset with spatial relation annotations and object features for image-to-text generation, built using 2,026 images from the PASCAL VOC2008 dataset. The dataset incorporates (i) the labelled object bounding boxes from VOC2008, (ii) geometrical, language and depth features for each object, and (iii) for each pair of objects in both orders, (a) the single best preposition and (b) the set of possible prepositions in the given language that describe the spatial relationship between the two objects. Compared to previous versions of the dataset, we have roughly doubled the size for French, and completely reannotated as well as increased the size of the English portion, providing single best prepositions for English for the first time. Furthermore, we have added explicit 3D depth features for objects. We are releasing our dataset for free reuse, along with evaluation tools to enable comparative evaluation.

Detection of spatial relations between objects in images is currently a popular subject in image description research. A range of different language and geometric object features have been used in this context, but methods have not so far used explicit information about the third dimension (depth), except when manually added to annotations. The lack of such information hampers detection of spatial relations that are inherently 3D. In this paper, we use a fully automatic method for creating a depth map of an image and derive several different object-level depth features from it which we add to an existing feature set to test the effect on spatial relation detection. We show that performance increases are obtained from adding depth features in all scenarios tested.

In this paper, we present the datasets used in the Shallow and Deep Tracks of the First Multilingual Surface Realisation Shared Task (SR’18). For the Shallow Track, data in ten languages has been released: Arabic, Czech, Dutch, English, Finnish, French, Italian, Portuguese, Russian and Spanish. For the Deep Track, data in three languages is made available: English, French and Spanish. We describe in detail how the datasets were derived from the Universal Dependencies V2.0, and report on an evaluation of the Deep Track input quality. In addition, we examine the motivation for, and likely usefulness of, deriving NLG inputs from annotations in resources originally developed for Natural Language Understanding (NLU), and assess whether the resulting inputs supply enough information of the right kind for the final stage in the NLG process.

We propose a shared task on multilingual Surface Realization, i.e., on mapping unordered and uninflected universal dependency trees to correctly ordered and inflected sentences in a number of languages. A second deeper input will be available in which, in addition, functional words, fine-grained PoS and morphological information will be removed from the input trees. The first shared task on Surface Realization was carried out in 2011 with a similar setup, with a focus on English. We think that it is time for relaunching such a shared task effort in view of the arrival of Universal Dependencies annotated treebanks for a large number of languages on the one hand, and the increasing dominance of Deep Learning, which proved to be a game changer for NLP, on the other hand.

Postmarketing surveillance (PMS) has the vital aim to monitor effects of drugs after release for use by the general population, but suffers from under-reporting and limited coverage. Automatic methods for detecting drug effect reports, especially for social media, could vastly increase the scope of PMS. Very few automatic PMS methods are currently available, in particular for the messy text types encountered on Twitter. In this paper we describe first results for developing PMS methods specifically for tweets. We describe the corpus of 125,669 tweets we have created and annotated to train and test the tools. We find that generic tools perform well for tweet-level language identification and tweet-level sentiment analysis (both 0.94 F1-Score). For detection of effect mentions we are able to achieve 0.87 F1-Score, while effect-level adverse-vs.-beneficial analysis proves harder with an F1-Score of 0.64. Among other things, our results indicate that MetaMap semantic types provide a very promising basis for identifying drug effect mentions in tweets.

Interactive systems have become an increasingly important type of application for deployment of NLG technology over recent years. At present, we do not yet have commonly agreed terminology or methodology for evaluating NLG within interactive systems. In this paper, we take steps towards addressing this gap by presenting a set of principles for designing new evaluations in our comparative evaluation methodology. We start with presenting a categorisation framework, giving an overview of different categories of evaluation measures, in order to provide standard terminology for categorising existing and new evaluation techniques. Background on existing evaluation methodologies for NLG and interactive systems is presented. The comparative evaluation methodology is presented. Finally, a methodology for comparative evaluation of NLG components embedded within interactive systems is presented in terms of the comparative evaluation methodology, using a specific task for illustrative purposes.

Starting in 2007, the field of natural language generation (NLG) has organised shared-task evaluation events every year, under the Generation Challenges umbrella. In the course of these shared tasks, a wealth of data has been created, along with associated task definitions and evaluation regimes. In other contexts too, sharable NLG data is now being created. In this paper, we describe the online repository that we have created as a one-stop resource for obtaining NLG task materials, both from Generation Challenges tasks and from other sources, where the set of materials provided for each task consists of (i) task definition, (ii) input and output data, (iii) evaluation software, (iv) documentation, and (v) publications reporting previous results.

In this paper we describe the LG-Eval toolkit for creating online language evaluation experiments. LG-Eval is the direct result of our work setting up and carrying out the human evaluation experiments in several of the Generation Challenges shared tasks. It provides tools for creating experiments with different kinds of rating tools, allocating items to evaluators, and collecting the evaluation scores.

Creating language resources is expensive and time-consuming, and this forms a bottleneck in the development of language technology, for less-studied non-European languages in particular. The recent internet phenomenon of crowd-sourcing offers a cost-effective and potentially fast way of overcoming such language resource acquisition bottlenecks. We present a methodology that takes as its input scanned documents of typed or hand-written text, and produces transcriptions of the text as its output. Instead of using Optical Character Recognition (OCR) technology, the methodology is game-based and produces such transcriptions as a by-product. The approach is intended particularly for languages for which language technology and resources are scarce and reliable OCR technology may not exist. It can be used in place of OCR for transcribing individual documents, or to create corpora of paired images and transcriptions required to train OCR tools. We present Minefield, a prototype implementation of the approach which is currently collecting Arabic transcriptions.