Current methods for evaluation of natural language generation models focus on measuring text quality but fail to probe the model creativity, i.e., its ability to generate novel but coherent text sequences not seen in the training corpus. We present the GenX tool which is designed to enable interactive exploration and explanation of natural language generation outputs with a focus on the detection of memorization. We demonstrate the utility of the tool on two domain-conditioned generation use cases - phishing emails and ACL abstracts.

Evaluation beyond aggregate performance metrics, e.g. F1-score, is crucial to both establish an appropriate level of trust in machine learning models and identify avenues for future model improvements. In this paper we demonstrate CrossCheck, an interactive capability for rapid cross-model comparison and reproducible error analysis. We describe the tool, discuss design and implementation details, and present three NLP use cases – named entity recognition, reading comprehension, and clickbait detection that show the benefits of using the tool for model evaluation. CrossCheck enables users to make informed decisions when choosing between multiple models, identify when the models are correct and for which examples, investigate whether the models are making the same mistakes as humans, evaluate models’ generalizability and highlight models’ limitations, strengths and weaknesses. Furthermore, CrossCheck is implemented as a Jupyter widget, which allows for rapid and convenient integration into existing model development workflows.

Traditional synonym recommendations often include ill-suited suggestions for writer’s specific contexts. We propose a simple approach for contextual synonym recommendation by combining existing human-curated thesauri, e.g. WordNet, with pre-trained language models. We evaluate our technique by curating a set of word-sentence pairs balanced across corpora and parts of speech, then annotating each word-sentence pair with the contextually appropriate set of synonyms. We found that basic language model approaches have higher precision. Approaches leveraging sentence context have higher recall. Overall, the latter contextual approach had the highest F-score.

We evaluate neural model robustness to adversarial attacks using different types of linguistic unit perturbations – character and word, and propose a new method for strategic sentence-level perturbations. We experiment with different amounts of perturbations to examine model confidence and misclassification rate, and contrast model performance with different embeddings BERT and ELMo on two benchmark datasets SQuAD and TriviaQA. We demonstrate how to improve model performance during an adversarial attack by using ensembles. Finally, we analyze factors that effect model behavior under adversarial attack, and develop a new model to predict errors during attacks. Our novel findings reveal that (a) unlike BERT, models that use ELMo embeddings are more susceptible to adversarial attacks, (b) unlike word and paraphrase, character perturbations affect the model the most but are most easily compensated for by adversarial training, (c) word perturbations lead to more high-confidence misclassifications compared to sentence- and character-level perturbations, (d) the type of question and model answer length (the longer the answer the more likely it is to be incorrect) is the most predictive of model errors in adversarial setting, and (e) conclusions about model behavior are dataset-specific.

With the increasing use of machine-learning driven algorithmic judgements, it is critical to develop models that are robust to evolving or manipulated inputs. We propose an extensive analysis of model robustness against linguistic variation in the setting of deceptive news detection, an important task in the context of misinformation spread online. We consider two prediction tasks and compare three state-of-the-art embeddings to highlight consistent trends in model performance, high confidence misclassifications, and high impact failures. By measuring the effectiveness of adversarial defense strategies and evaluating model susceptibility to adversarial attacks using character- and word-perturbed text, we find that character or mixed ensemble models are the most effective defenses and that character perturbation-based attack tactics are more successful.

Evaluating model robustness is critical when developing trustworthy models not only to gain deeper understanding of model behavior, strengths, and weaknesses, but also to develop future models that are generalizable and robust across expected environments a model may encounter in deployment. In this paper, we present a framework for measuring model robustness for an important but difficult text classification task – deceptive news detection. We evaluate model robustness to out-of-domain data, modality-specific features, and languages other than English. Our investigation focuses on three type of models: LSTM models trained on multiple datasets (Cross-Domain), several fusion LSTM models trained with images and text and evaluated with three state-of-the-art embeddings, BERT ELMo, and GloVe (Cross-Modality), and character-level CNN models trained on multiple languages (Cross-Language). Our analyses reveal a significant drop in performance when testing neural models on out-of-domain data and non-English languages that may be mitigated using diverse training data. We find that with additional image content as input, ELMo embeddings yield significantly fewer errors compared to BERT or GLoVe. Most importantly, this work not only carefully analyzes deception model robustness but also provides a framework of these analyses that can be applied to new models or extended datasets in the future.

Language in social media is a dynamic system, constantly evolving and adapting, with words and concepts rapidly emerging, disappearing, and changing their meaning. These changes can be estimated using word representations in context, over time and across locations. A number of methods have been proposed to track these spatiotemporal changes but no general method exists to evaluate the quality of these representations. Previous work largely focused on qualitative evaluation, which we improve by proposing a set of visualizations that highlight changes in text representation over both space and time. We demonstrate usefulness of novel spatiotemporal representations to explore and characterize specific aspects of the corpus of tweets collected from European countries over a two-week period centered around the terrorist attacks in Brussels in March 2016. In addition, we quantitatively evaluate spatiotemporal representations by feeding them into a downstream classification task – event type prediction. Thus, our work is the first to provide both intrinsic (qualitative) and extrinsic (quantitative) evaluation of text representations for spatiotemporal trends.