In this tutorial, we focus on text-to-text generation, a class of natural language generation (NLG) tasks, that takes a piece of text as input and then generates a revision that is improved according to some specific criteria (e.g., readability or linguistic styles), while largely retaining the original meaning and the length of the text. This includes many useful applications, such as text simplification, paraphrase generation, style transfer, etc. In contrast to text summarization and open-ended text completion (e.g., story), the text-to-text generation tasks we discuss in this tutorial are more constrained in terms of semantic consistency and targeted language styles. This level of control makes these tasks ideal testbeds for studying the ability of models to generate text that is both semantically adequate and stylistically appropriate. Moreover, these tasks are interesting from a technical standpoint, as they require complex combinations of lexical and syntactical transformations, stylistic control, and adherence to factual knowledge, – all at once. With a special focus on text simplification and revision, this tutorial aims to provide an overview of the state-of-the-art natural language generation research from four major aspects – Data, Models, Human-AI Collaboration, and Evaluation – and to discuss and showcase a few significant and recent advances: (1) the use of non-retrogressive approaches; (2) the shift from fine-tuning to prompting with large language models; (3) the development of new learnable metric and fine-grained human evaluation framework; (4) a growing body of studies and datasets on non-English languages; (5) the rise of HCI+NLP+Accessibility interdisciplinary research to create real-world writing assistant systems.

Multilingual generation from Abstract Meaning Representations (AMRs) verbalises AMRs into multiple languages. Previous work has focused on high- and medium-resource languages relying on large amounts of training data. In this work, we consider both high- and low-resource languages capping training data size at the lower bound set by our low-resource languages i.e. 31K. We propose a straightforward technique to enhance results on low-resource while preserving performance on high-resource languages. We iteratively refine a multilingua model to a set of monolingual models using Low-Rank Adaptation with a training curriculum based on a tree structure; this permits investigating how the languages used at each iteration impact generation performance on high and low-resource languages. We show an improvement over both mono and multilingual approaches. Comparing different ways of grouping languages at each iteration step we find two working configurations: grouping related languages which promotes transfer, or grouping distant languages which facilitates regularisation

While the WebNLG dataset has prompted much research on generation from knowledge graphs, little work has examined how well models trained on the WebNLG data generalise to unseen data and work has mostly been focused on English. In this paper, we introduce novel benchmarks for both English and Russian which contain various ratios of unseen entities and properties. These benchmarks also differ from WebNLG in that some of the graphs stem from Wikidata rather than DBpedia. Evaluating various models for English and Russian on these benchmarks shows a strong decrease in performance while a qualitative analysis highlights the various types of errors induced by non i.i.d data.

We present KGConv, a large corpus of 71k English conversations where each question-answer pair is grounded in a Wikidata fact. Conversations contain on average 8.6 questions and for each Wikidata fact, we provide multiple variants (12 on average) of the corresponding question using templates, human annotations, hand-crafted rules and a question rewriting neural model. We provide baselines for the task of Knowledge-Based, Conversational Question Generation. KGConv can further be used for other generation and analysis tasks such as single-turn question generation from Wikidata triples, question rewriting, question answering from conversation or from knowledge graphs and quiz generation.

Text simplification intends to make a text easier to read while preserving its core meaning. Intuitively and as shown in previous works, these two dimensions (simplification and meaning preservation) are often-times inversely correlated. An overly conservative text will fail to simplify sufficiently, whereas extreme simplification will degrade meaning preservation. Yet, popular evaluation metrics either aggregate meaning preservation and simplification into a single score (SARI, LENS), or target meaning preservation alone (BERTScore, QuestEval). Moreover, these metrics usually require a set of references and most previous work has only focused on sentence-level simplification. In this paper, we focus on the evaluation of document-level text simplification and compare existing models using distinct metrics for meaning preservation and simplification. We leverage existing metrics from similar tasks and introduce a reference-less metric variant for simplicity, showing that models are mostly biased towards either simplification or meaning preservation, seldom performing well on both dimensions. Making use of the fact that the metrics we use are all reference-less, we also investigate the performance of existing models when applied to unseen data (where reference simplifications are unavailable).

Most existing work on text simplification is limited to sentence-level inputs, with attempts to iteratively apply these approaches to document-level simplification failing to coherently preserve the discourse structure of the document. We hypothesise that by providing a high-level view of the target document, a simplification plan might help to guide generation. Building upon previous work on controlled, sentence-level simplification, we view a plan as a sequence of labels, each describing one of four sentence-level simplification operations (copy, rephrase, split, or delete). We propose a planning model that labels each sentence in the input document while considering both its context (a window of surrounding sentences) and its internal structure (a token-level representation). Experiments on two simplification benchmarks (Newsela-auto and Wiki-auto) show that our model outperforms strong baselines both on the planning task and when used to guide document-level simplification models.

Automatic evaluation for sentence simplification remains a challenging problem. Most popular evaluation metrics require multiple high-quality references – something not readily available for simplification – which makes it difficult to test performance on unseen domains. Furthermore, most existing metrics conflate simplicity with correlated attributes such as fluency or meaning preservation. We propose a new learned evaluation metric — SLE — which focuses on simplicity, outperforming almost all existing metrics in terms of correlation with human judgements.

To date, most work on text simplification has focused on sentence-level inputs. Early attempts at document simplification merely applied these approaches iteratively over the sentences of a document. However, this fails to coherently preserve the discourse structure, leading to suboptimal output quality. Recently, strategies from controllable simplification have been leveraged to achieve state-of-the-art results on document simplification by first generating a document-level plan (a sequence of sentence-level simplification operations) and using this plan to guide sentence-level simplification downstream. However, this is still limited in that the simplification model has no direct access to the local inter-sentence document context, likely having a negative impact on surface realisation. We explore various systems that use document context within the simplification process itself, either by iterating over larger text units or by extending the system architecture to attend over a high-level representation of document context. In doing so, we achieve state-of-the-art performance on the document simplification task, even when not relying on plan-guidance. Further, we investigate the performance and efficiency tradeoffs of system variants and make suggestions of when each should be preferred.

The ability to bridge Question Generation (QG) and Question Answering (QA) across structured and unstructured modalities has the potential for aiding different NLP applications. One key application is in QA-based methods that have recently been shown to be useful for automatically evaluating Natural Language (NL) texts generated from Knowledge Graphs (KG). While methods have been proposed for QG-QA across these modalities, these efforts have been in English only; in this work, we bring multilinguality (Brazilian Portuguese and Russian) to multimodal (KG and NL) QG-QA. Using synthetic data generation and machine translation to produce QG-QA data that is aligned between graph and text, we are able to train multimodal, multi-task models that can perform multimodal QG and QA in Portuguese and Russian. We show that our approach outperforms a baseline which is derived from previous work on English and adapted to handle these two languages.

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.

Generating factual, long-form text such as Wikipedia articles raises three key challenges: how to gather relevant evidence, how to structure information into well-formed text, and how to ensure that the generated text is factually correct. We address these by developing a model for English text that uses a retrieval mechanism to identify relevant supporting information on the web and a cache-based pre-trained encoder-decoder to generate long-form biographies section by section, including citation information. To assess the impact of available web evidence on the output text, we compare the performance of our approach when generating biographies about women (for which less information is available on the web) vs. biographies generally. To this end, we curate a dataset of 1,500 biographies about women. We analyze our generated text to understand how differences in available web evidence data affect generation. We evaluate the factuality, fluency, and quality of the generated texts using automatic metrics and human evaluation. We hope that these techniques can be used as a starting point for human writers, to aid in reducing the complexity inherent in the creation of long-form, factual text.

Different types of transformations have been used to model sentence simplification ranging from mainly local operations such as phrasal or lexical rewriting, deletion and re-ordering to the more global affecting the whole input sentence such as sentence rephrasing, copying and splitting. In this paper, we propose a novel approach to sentence simplification which encompasses four global operations: whether to rephrase or copy and whether to split based on syntactic or discourse structure. We create a novel dataset that can be used to train highly accurate classification systems for these four operations. We propose a controllable-simplification model that tailors simplifications to these operations and show that it outperforms both end-to-end, non-controllable approaches and previous controllable approaches.

In the medical field, we have seen the emergence of health-bots that interact with patients to gather data and track their state. One of the downstream application is automatic questionnaire filling, where the content of the dialog is used to automatically fill a pre-defined medical questionnaire. Previous work has shown that answering questions from the dialog context can successfully be cast as a Natural Language Inference (NLI) task and therefore benefit from current pre-trained NLI models. However, NLI models have mostly been trained on text rather than dialogs, which may have an influence on their performance. In this paper, we study the influence of content transformation and content selection on the questionnaire filling task. Our results demonstrate that dialog pre-processing can significantly improve the performance of zero-shot questionnaire filling models which take health-bots dialogs as input.

Question generation from knowledge bases (or knowledge base question generation, KBQG) is the task of generating questions from structured database information, typically in the form of triples representing facts. To handle rare entities and generalize to unseen properties, previous work on KBQG resorted to extensive, often ad-hoc pre- and post-processing of the input triple. We revisit KBQG – using pre training, a new (triple, question) dataset and taking question type into account – and show that our approach outperforms previous work both in a standard and in a zero-shot setting. We also show that the extended KBQG dataset (also helpful for knowledge base question answering) we provide allows not only for better coverage in terms of knowledge base (KB) properties but also for increased output variability in that it permits the generation of multiple questions from the same KB triple.

Sentence splitting involves the segmentation of a sentence into two or more shorter sentences. It is a key component of sentence simplification, has been shown to help human comprehension and is a useful preprocessing step for NLP tasks such as summarisation and relation extraction. While several methods and datasets have been proposed for developing sentence splitting models, little attention has been paid to how sentence splitting interacts with discourse structure. In this work, we focus on cases where the input text contains a discourse connective, which we refer to as discourse-based sentence splitting. We create synthetic and organic datasets for discourse-based splitting and explore different ways of combining these datasets using different model architectures. We show that pipeline models which use discourse structure to mediate sentence splitting outperform end-to-end models in learning the various ways of expressing a discourse relation but generate text that is less grammatical; that large scale synthetic data provides a better basis for learning than smaller scale organic data; and that training on discourse-focused, rather than on general sentence splitting data provides a better basis for discourse splitting.

While powerful pre-trained language models have improved the fluency of text generation models, semantic adequacy -the ability to generate text that is semantically faithful to the input- remains an unsolved issue. In this paper, we introduce a novel automatic evaluation metric, Entity-Based Semantic Adequacy, which can be used to assess to what extent generation models that verbalise RDF (Resource Description Framework) graphs produce text that contains mentions of the entities occurring in the RDF input. This is important as RDF subject and object entities make up 2/3 of the input. We use our metric to compare 25 models from the WebNLG Shared Tasks and we examine correlation with results from human evaluations of semantic adequacy. We show that while our metric correlates with human evaluation scores, this correlation varies with the specifics of the human evaluation setup. This suggests that in order to measure the entity-based adequacy of generated texts, an automatic metric such as the one proposed here might be more reliable, as less subjective and more focused on correct verbalisation of the input, than human evaluation measures.

We focus on dialog models in the context of clinical studies where the goal is to help gather, in addition to the close information collected based on a questionnaire, serendipitous information that is medically relevant. To promote user engagement and address this dual goal (collecting both a predefined set of data points and more informal information about the state of the patients), we introduce an ensemble model made of three bots: a task-based, a follow-up and a social bot. We introduce a generic method for developing follow-up bots. We compare different ensemble configurations and we show that the combination of the three bots (i) provides a better basis for collecting information than just the information seeking bot and (ii) collects information in a more user-friendly, more efficient manner that an ensemble model combining the information seeking and the social bot.

Various machine learning tasks can benefit from access to external information of different modalities, such as text and images. Recent work has focused on learning architectures with large memories capable of storing this knowledge. We propose augmenting generative Transformer neural networks with KNN-based Information Fetching (KIF) modules. Each KIF module learns a read operation to access fixed external knowledge. We apply these modules to generative dialog modeling, a challenging task where information must be flexibly retrieved and incorporated to maintain the topic and flow of conversation. We demonstrate the effectiveness of our approach by identifying relevant knowledge required for knowledgeable but engaging dialog from Wikipedia, images, and human-written dialog utterances, and show that leveraging this retrieved information improves model performance, measured by automatic and human evaluation.

The metrics standardly used to evaluate Natural Language Generation (NLG) models, such as BLEU or METEOR, fail to provide information on which linguistic factors impact performance. Focusing on Surface Realization (SR), the task of converting an unordered dependency tree into a well-formed sentence, we propose a framework for error analysis which permits identifying which features of the input affect the models’ results. This framework consists of two main components: (i) correlation analyses between a wide range of syntactic metrics and standard performance metrics and (ii) a set of techniques to automatically identify syntactic constructs that often co-occur with low performance scores. We demonstrate the advantages of our framework by performing error analysis on the results of 174 system runs submitted to the Multilingual SR shared tasks; we show that dependency edge accuracy correlate with automatic metrics thereby providing a more interpretable basis for evaluation; and we suggest ways in which our framework could be used to improve models and data. The framework is available in the form of a toolkit which can be used both by campaign organizers to provide detailed, linguistically interpretable feedback on the state of the art in multilingual SR, and by individual researchers to improve models and datasets.1

A key bottleneck for developing dialog models is the lack of adequate training data. Due to privacy issues, dialog data is even scarcer in the health domain. We propose a novel method for creating dialog corpora which we apply to create doctor-patient interaction data. We use this data to learn both a generation and a hybrid classification/retrieval model and find that the generation model consistently outperforms the hybrid model. We show that our data creation method has several advantages. Not only does it allow for the semi-automatic creation of large quantities of training data. It also provides a natural way of guiding learning and a novel method for assessing the quality of human-machine interactions.

Generating text from structured data is challenging because it requires bridging the gap between (i) structure and natural language (NL) and (ii) semantically underspecified input and fully specified NL output. Multilingual generation brings in an additional challenge: that of generating into languages with varied word order and morphological properties. In this work, we focus on Abstract Meaning Representations (AMRs) as structured input, where previous research has overwhelmingly focused on generating only into English. We leverage advances in cross-lingual embeddings, pretraining, and multilingual models to create multilingual AMR-to-text models that generate in twenty one different languages. Our multilingual models surpass baselines that generate into one language in eighteen languages, based on automatic metrics. We analyze the ability of our multilingual models to accurately capture morphology and word order using human evaluation, and find that native speakers judge our generations to be fluent.

End-to-end encoder-decoder approaches to data-to-text generation are often black boxes whose predictions are difficult to explain. Breaking up the end-to-end model into sub-modules is a natural way to address this problem. The traditional pre-neural Natural Language Generation (NLG) pipeline provides a framework for breaking up the end-to-end encoder-decoder. We survey recent papers that integrate traditional NLG submodules in neural approaches and analyse their explainability. Our survey is a first step towards building explainable neural NLG models.

Recent graph-to-text models generate text from graph-based data using either global or local aggregation to learn node representations. Global node encoding allows explicit communication between two distant nodes, thereby neglecting graph topology as all nodes are directly connected. In contrast, local node encoding considers the relations between neighbor nodes capturing the graph structure, but it can fail to capture long-range relations. In this work, we gather both encoding strategies, proposing novel neural models that encode an input graph combining both global and local node contexts, in order to learn better contextualized node embeddings. In our experiments, we demonstrate that our approaches lead to significant improvements on two graph-to-text datasets achieving BLEU scores of 18.01 on the AGENDA dataset, and 63.69 on the WebNLG dataset for seen categories, outperforming state-of-the-art models by 3.7 and 3.1 points, respectively.1

The RDF-to-text task has recently gained substantial attention due to the continuous growth of RDF knowledge graphs in number and size. Recent studies have focused on systematically comparing RDF-to-text approaches on benchmarking datasets such as WebNLG. Although some evaluation tools have already been proposed for text generation, none of the existing solutions abides by the Findability, Accessibility, Interoperability, and Reusability (FAIR) principles and involves RDF data for the knowledge extraction task. In this paper, we present BENG, a FAIR benchmarking platform for Natural Language Generation (NLG) and Knowledge Extraction systems with focus on RDF data. BENG builds upon the successful benchmarking platform GERBIL, is opensource and is publicly available along with the data it contains.

WebNLG+ offers two challenges: (i) mapping sets of RDF triples to English or Russian text (generation) and (ii) converting English or Russian text to sets of RDF triples (semantic parsing). Compared to the eponymous WebNLG challenge, WebNLG+ provides an extended dataset that enable the training, evaluation, and comparison of microplanners and semantic parsers. In this paper, we present the results of the generation and semantic parsing task for both English and Russian and provide a brief description of the participating systems.

Surface realisation (SR) maps a meaning representation to a sentence and can be viewed as consisting of three subtasks: word ordering, morphological inflection and contraction generation (e.g., clitic attachment in Portuguese or elision in French). We propose a modular approach to surface realisation which models each of these components separately, and evaluate our approach on the 10 languages covered by the SR’18 Surface Realisation Shared Task shallow track. We provide a detailed evaluation of how word order, morphological realisation and contractions are handled by the model and an analysis of the differences in word ordering performance across languages.

Generating text from graph-based data, such as Abstract Meaning Representation (AMR), is a challenging task due to the inherent difficulty in how to properly encode the structure of a graph with labeled edges. To address this difficulty, we propose a novel graph-to-sequence model that encodes different but complementary perspectives of the structural information contained in the AMR graph. The model learns parallel top-down and bottom-up representations of nodes capturing contrasting views of the graph. We also investigate the use of different node message passing strategies, employing different state-of-the-art graph encoders to compute node representations based on incoming and outgoing perspectives. In our experiments, we demonstrate that the dual graph representation leads to improvements in AMR-to-text generation, achieving state-of-the-art results on two AMR datasets

Query-based open-domain NLP tasks require information synthesis from long and diverse web results. Current approaches extractively select portions of web text as input to Sequence-to-Sequence models using methods such as TF-IDF ranking. We propose constructing a local graph structured knowledge base for each query, which compresses the web search information and reduces redundancy. We show that by linearizing the graph into a structured input sequence, models can encode the graph representations within a standard Sequence-to-Sequence setting. For two generative tasks with very long text input, long-form question answering and multi-document summarization, feeding graph representations as input can achieve better performance than using retrieved text portions.

This paper presents the LORIA / Lorraine University submission at the Multilingual Surface Realisation shared task 2019 for the shallow track. We outline our approach and evaluate it on 11 languages covered by the shared task. We provide a separate evaluation of each component of our pipeline, concluding on some difficulties and suggesting directions for future work.

In this paper, we propose an approach for semi-automatically creating a data-to-text (D2T) corpus for Russian that can be used to learn a D2T natural language generation model. An error analysis of the output of an English-to-Russian neural machine translation system shows that 80% of the automatically translated sentences contain an error and that 53% of all translation errors bear on named entities (NE). We therefore focus on named entities and introduce two post-editing techniques for correcting wrongly translated NEs.

Surface realisation (SR) consists in generating a text from a meaning representations (MR). In this paper, we introduce a new parallel dataset of deep meaning representations (MR) and French sentences and we present a novel method for MR-to-text generation which seeks to generalise by abstracting away from lexical content. Most current work on natural language generation focuses on generating text that matches a reference using BLEU as evaluation criteria. In this paper, we additionally consider the model’s ability to reintroduce the function words that are absent from the deep input meaning representations. We show that our approach increases both BLEU score and the scores used to assess function words generation.

End-to-end neural approaches have achieved state-of-the-art performance in many natural language processing (NLP) tasks. Yet, they often lack transparency of the underlying decision-making process, hindering error analysis and certain model improvements. In this work, we revisit the binary linearization approach to surface realization, which exhibits more interpretable behavior, but was falling short in terms of prediction accuracy. We show how enriching the training data to better capture word order constraints almost doubles the performance of the system. We further demonstrate that encoding both local and global prediction contexts yields another considerable performance boost. With the proposed modifications, the system which ranked low in the latest shared task on multilingual surface realization now achieves best results in five out of ten languages, while being on par with the state-of-the-art approaches in others.

We study the automatic generation of syntactic paraphrases using four different models for generation: data-to-text generation, text-to-text generation, text reduction and text expansion, We derive training data for each of these tasks from the WebNLG dataset and we show (i) that conditioning generation on syntactic constraints effectively permits the generation of syntactically distinct paraphrases for the same input and (ii) that exploiting different types of input (data, text or data+text) further increases the number of distinct paraphrases that can be generated for a given input.

Text production is a key component of many NLP applications. In data-driven approaches, it is used for instance, to generate dialogue turns from dialogue moves, to verbalise the content of Knowledge bases or to generate natural English sentences from rich linguistic representations, such as dependency trees or Abstract Meaning Representations. In text-driven methods on the other hand, text production is at work in sentence compression, sentence fusion, paraphrasing, sentence (or text) simplification, text summarisation and end-to-end dialogue systems. Following the success of encoder-decoder models in modeling sequence-rewriting tasks such as machine translation, deep learning models have successfully been applied to the various text production tasks. In this tutorial, we will cover the fundamentals and the state-of-the-art research on neural models for text production. Each text production task raises a slightly different communication goal (e.g, how to take the dialogue context into account when producing a dialogue turn; how to detect and merge relevant information when summarising a text; or how to produce a well-formed text that correctly capture the information contained in some input data in the case of data-to-text generation). We will outline the constraints specific to each subtasks and examine how the existing neural models account for them.

Neural approaches to data-to-text generation generally handle rare input items using either delexicalisation or a copy mechanism. We investigate the relative impact of these two methods on two datasets (E2E and WebNLG) and using two evaluation settings. We show (i) that rare items strongly impact performance; (ii) that combining delexicalisation and copying yields the strongest improvement; (iii) that copying underperforms for rare and unseen items and (iv) that the impact of these two mechanisms greatly varies depending on how the dataset is constructed and on how it is split into train, dev and test.

We propose a new sentence simplification task (Split-and-Rephrase) where the aim is to split a complex sentence into a meaning preserving sequence of shorter sentences. Like sentence simplification, splitting-and-rephrasing has the potential of benefiting both natural language processing and societal applications. Because shorter sentences are generally better processed by NLP systems, it could be used as a preprocessing step which facilitates and improves the performance of parsers, semantic role labellers and machine translation systems. It should also be of use for people with reading disabilities because it allows the conversion of longer sentences into shorter ones. This paper makes two contributions towards this new task. First, we create and make available a benchmark consisting of 1,066,115 tuples mapping a single complex sentence to a sequence of sentences expressing the same meaning. Second, we propose five models (vanilla sequence-to-sequence to semantically-motivated models) to understand the difficulty of the proposed task.

Although there has been much work in recent years on data-driven natural language generation, little attention has been paid to the fine-grained interactions that arise during microplanning between aggregation, surface realization, and sentence segmentation. In this article, we propose a hybrid symbolic/statistical approach to jointly model the constraints regulating these interactions. Our approach integrates a small handwritten grammar, a statistical hypertagger, and a surface realization algorithm. It is applied to the verbalization of knowledge base queries and tested on 13 knowledge bases to demonstrate domain independence. We evaluate our approach in several ways. A quantitative analysis shows that the hybrid approach outperforms a purely symbolic approach in terms of both speed and coverage. Results from a human study indicate that users find the output of this hybrid statistic/symbolic system more fluent than both a template-based and a purely symbolic grammar-based approach. Finally, we illustrate by means of examples that our approach can account for various factors impacting aggregation, sentence segmentation, and surface realization.

In this paper, we present a novel framework for semi-automatically creating linguistically challenging micro-planning data-to-text corpora from existing Knowledge Bases. Because our method pairs data of varying size and shape with texts ranging from simple clauses to short texts, a dataset created using this framework provides a challenging benchmark for microplanning. Another feature of this framework is that it can be applied to any large scale knowledge base and can therefore be used to train and learn KB verbalisers. We apply our framework to DBpedia data and compare the resulting dataset with Wen et al. 2016’s. We show that while Wen et al.’s dataset is more than twice larger than ours, it is less diverse both in terms of input and in terms of text. We thus propose our corpus generation framework as a novel method for creating challenging data sets from which NLG models can be learned which are capable of handling the complex interactions occurring during in micro-planning between lexicalisation, aggregation, surface realisation, referring expression generation and sentence segmentation. To encourage researchers to take up this challenge, we made available a dataset of 21,855 data/text pairs created using this framework in the context of the WebNLG shared task.

The WebNLG challenge consists in mapping sets of RDF triples to text. It provides a common benchmark on which to train, evaluate and compare “microplanners”, i.e. generation systems that verbalise a given content by making a range of complex interacting choices including referring expression generation, aggregation, lexicalisation, surface realisation and sentence segmentation. In this paper, we introduce the microplanning task, describe data preparation, introduce our evaluation methodology, analyse participant results and provide a brief description of the participating systems.

A generation system can only be as good as the data it is trained on. In this short paper, we propose a methodology for analysing data-to-text corpora used for training Natural Language Generation (NLG) systems. We apply this methodology to three existing benchmarks. We conclude by eliciting a set of criteria for the creation of a data-to-text benchmark which could help better support the development, evaluation and comparison of linguistically sophisticated data-to-text generators.

In Natural Language Generation (NLG), one important limitation is the lack of common benchmarks on which to train, evaluate and compare data-to-text generators. In this paper, we make one step in that direction and introduce a method for automatically creating an arbitrary large repertoire of data units that could serve as input for generation. Using both automated metrics and a human evaluation, we show that the data units produced by our method are both diverse and coherent.

We describe the acquisition of a dialog corpus for French based on multi-task human-machine interactions in a serious game setting. We present a tool for data collection that is configurable for multiple games; describe the data collected using this tool and the annotation schema used to annotate it; and report on the results obtained when training a classifier on the annotated data to associate each player turn with a dialog move usable by a rule based dialog manager. The collected data consists of approximately 1250 dialogs, 10454 utterances and 168509 words and will be made freely available to academic and nonprofit research.

There has been much debate, both theoretical and practical, on how to link ontologies and lexicons in natural language processing (NLP) applications. In this paper, we focus on an application in which lexicon and ontology are used to generate teaching material. We briefly describe the application (a serious game for language learning). We then zoom in on the representation and interlinking of the lexicon and of the ontology. We show how the use of existing standards and of good practice principles facilitates the design of our resources while satisfying the expressivity requirements set by natural language generation.

Ce travail prend place dans le cadre plus général du développement d’une plate-forme d’analyse syntaxique du français parlé. Nous décrivons la conception d’un modèle automatique pour résoudre le lien anaphorique présent dans les dislocations à gauche dans un corpus de français parlé radiophonique. La détection de ces structures devrait permettre à terme d’améliorer notre analyseur syntaxique en enrichissant les informations prises en compte dans nos modèles automatiques. La résolution du lien anaphorique est réalisée en deux étapes : un premier niveau à base de règles filtre les configurations candidates, et un second niveau s’appuie sur un modèle appris selon le critère du maximum d’entropie. Une évaluation expérimentale réalisée par validation croisée sur un corpus annoté manuellement donne une F-mesure de l’ordre de 40%.

Previous work has shown that large scale subcategorisation lexicons could be extracted from parsed corpora with reasonably high precision. In this paper, we apply a standard extraction procedure to a 100 millions words parsed corpus of french and obtain rather poor results. We investigate different factors likely to improve performance such as in particular, the specific extraction procedure and the parser used; the size of the input corpus; and the type of frames learned. We try out different ways of interleaving the output of several parsers with the lexicon extraction process and show that none of them improves the results. Conversely, we show that increasing the size of the input corpus and modifying the extraction procedure to better differentiate prepositional arguments from prepositional modifiers improves performance. In conclusion, we suggest that a more sophisticated approach to parser combination and better probabilistic models of the various types of prepositional objects in French are likely ways to get better results.

We focus on textual entailments mediated by syntax and propose a new methodology to evaluate textual entailment recognition systems on such data. The main idea is to generate a syntactically annotated corpus of pairs of (non-)entailments and to use error mining methodology from the parsing field to identify the most likely sources of errors. To generate the evaluation corpus we use a template based generation approach where sentences, semantic representations and syntactic annotations are all created at the same time. Furthermore, we adapt the error mining methodology initially proposed for parsing to the field of textual entailment. To illustrate the approach, we apply the proposed methodology to the Afazio RTE system (an hybrid system focusing on syntactic entailment) and show how it permits identifying the most likely sources of errors made by this system on a testsuite of 10 000 (non-)entailment pairs which is balanced in term of (non-)entailment and in term of syntactic annotations.

Cet article décrit une méthodologie visant la réalisation d’une ressource sémantique en français centrée sur la synonymie. De manière complémentaire aux travaux existants, la méthode proposée n’a pas seulement pour objectif d’établir des liens de synonymie entre lexèmes, mais également d’apparier les sens possibles d’un lexème avec les ensembles de synonymes appropriés. En pratique, les sens possibles des lexèmes proviennent des définitions du TLFi et les synonymes de cinq dictionnaires accessibles à l’ATILF. Pour évaluer la méthode d’appariement entre sens d’un lexème et ensemble de synonymes, une ressource de référence a été réalisée pour 27 verbes du français par quatre lexicographes qui ont spécifié manuellement l’association entre verbe, sens (définition TLFi) et ensemble de synonymes. Relativement à ce standard étalon, la méthode d’appariement affiche une F-mesure de 0.706 lorsque l’ensemble des paramètres est pris en compte, notamment la distinction pronominal / non-pronominal pour les verbes du français et de 0.602 sans cette distinction.

Nous présentons un système de normalisation de la variation syntaxique qui permet de mieux reconnaître la relation d’implication textuelle entre deux phrases. Le système est évalué sur une suite de tests comportant 2 520 paires test et les résultats montrent un gain en précision par rapport à un système de base variant entre 29.8 et 78.5 points la complexité des cas considérés.

Recently, most of the research in NLP has concentrated on the creation of applications coping with textual entailment. However, there still exist very few resources for the evaluation of such applications. We argue that the reason for this resides not only in the novelty of the research field but also and mainly in the difficulty of defining the linguistic phenomena which are responsible for inference. As the TSNLP project has shown test suites provide optimal diagnostic and evaluation tools for NLP applications, as contrary to text corpora they provide a deep insight in the linguistic phenomena allowing control over the data. Thus in this paper, we present a test suite specifically developed for studying inference problems shown by English adjectives. The construction of the test suite is based on the deep linguistic analysis and following classification of entailment patterns of adjectives and follows the TSNLP guidelines on linguistic databases providing a clear coverage, systematic annotation of inference tasks, large reusability and simple maintenance. With the design of this test suite we aim at creating a resource supporting the evaluation of computational systems handling natural language inference and in particular at providing a benchmark against which to evaluate and compare existing semantic analysers.

Dans cet article, nous présentons une architecture logicielle libre et ouverte pour le développement de grammaires d’arbres adjoints à portée sémantique. Cette architecture utilise un compilateur de métagrammaires afin de faciliter l’extension et la maintenance de la grammaire, et intègre un module de construction sémantique permettant de vérifier la couverture aussi bien syntaxique que sémantique de la grammaire. Ce module utilise un analyseur syntaxique tabulaire généré automatiquement à partir de la grammaire par le système DyALog. Nous présentons également les résultats de l’évaluation d’une grammaire du français développée au moyen de cette architecture.

SYNLEX est un lexique syntaxique extrait semi-automatiquement des tables du LADL. Comme les autres lexiques syntaxiques du français disponibles et utilisables pour le TAL (LEFFF, DICOVALENCE), il est incomplet et n’a pas fait l’objet d’une évaluation permettant de déterminer son rappel et sa précision par rapport à un lexique de référence. Nous présentons une approche qui permet de combler au moins partiellement ces lacunes. L’approche s’appuie sur les méthodes mises au point en acquisition automatique de lexique. Un lexique syntaxique distinct de SYNLEX est acquis à partir d’un corpus de 82 millions de mots puis utilisé pour valider et compléter SYNLEX. Le rappel et la précision de cette version améliorée de SYNLEX sont ensuite calculés par rapport à un lexique de référence extrait de DICOVALENCE.

En génération, un réalisateur de surface a pour fonction de produire, à partir d’une représentation conceptuelle donnée, une phrase grammaticale. Les réalisateur existants soit utilisent une grammaire réversible et des méthodes statistiques pour déterminer parmi l’ensemble des sorties produites la plus plausible ; soit utilisent des grammaires spécialisées pour la génération et des méthodes symboliques pour déterminer la paraphrase la plus appropriée à un contexte de génération donné. Dans cet article, nous présentons GENI, un réalisateur de surface basé sur une grammaire d’arbres adjoints pour le français qui réconcilie les deux approches en combinant une grammaire réversible avec une sélection symbolique des paraphrases.

Les tables du LADL (Laboratoire d’Automatique Documentaire et Linguistique) contiennent des données électroniques extensives sur les propriétés morphosyntaxiques et syntaxiques des foncteurs syntaxiques du français (verbes, noms, adjectifs). Ces données, dont on sait qu’elles sont nécessaires pour le bon fonctionnement des systèmes de traitement automatique des langues, ne sont cependant que peu utilisées par les systèmes actuels. Dans cet article, nous identifions les raisons de cette lacune et nous proposons une méthode de conversion des tables vers un format mieux approprié au traitement automatique des langues.

Dans cet article, nous considérons un formalisme linguistique pour lequel l’intégration d’information sémantique dans une grammaire à large couverture n’a pas encore été réalisée à savoir, les grammaires d’arbres adjoints (Tree Adjoining Grammar ou TAG). Nous proposons une méthode permettant cette intégration et décrivons sa mise en oeuvre dans une grammaire noyau pour le français. Nous montrons en particulier que le formalisme de spécification utilisé, XMG, (Duchier et al., 2004) permet une factorisation importante des données sémantiques facilitant ainsi le développement, la maintenance et le déboggage de la grammaire.