Recently, large multi-modal models (LMMs) have emerged with the capacity to perform vision tasks such as captioning and visual question answering (VQA) with unprecedented accuracy. Applications such as helping the blind or visually impaired have a critical need for precise answers. It is specially important for models to be well calibrated and be able to quantify their uncertainty in order to selectively decide when to answer and when to abstain or ask for clarifications. We perform the first in-depth analysis of calibration methods and metrics for VQA with in-context learning LMMs. Studying VQA on two answerability benchmarks, we show that the likelihood score of visually grounded models is better calibrated than in their text-only counterparts for in-context learning, where sampling based methods are generally superior, but no clear winner arises. We propose Avg BLEU, a calibration score combining the benefits of both sampling and likelihood methods across modalities.

Warning: This paper contains explicit statements of offensive stereotypes which may be upsetting The study of bias, fairness and social impact in Natural Language Processing (NLP) lacks resources in languages other than English. Our objective is to support the evaluation of bias in language models in a multilingual setting. We use stereotypes across nine types of biases to build a corpus containing contrasting sentence pairs, one sentence that presents a stereotype concerning an underadvantaged group and another minimally changed sentence, concerning a matching advantaged group. We build on the French CrowS-Pairs corpus and guidelines to provide translations of the existing material into seven additional languages. In total, we produce 11,139 new sentence pairs that cover stereotypes dealing with nine types of biases in seven cultural contexts. We use the final resource for the evaluation of relevant monolingual and multilingual masked language models. We find that language models in all languages favor sentences that express stereotypes in most bias categories. The process of creating a resource that covers a wide range of language types and cultural settings highlights the difficulty of bias evaluation, in particular comparability across languages and contexts.

Approaches to bias assessment usually require such technical skills that, by design, they leave discrimination experts out. In this paper we present EDIA, a tool that facilitates that experts in discrimination explore social biases in word embeddings and masked language models. Experts can then characterize those biases so that their presence can be assessed more systematically, and actions can be planned to address them. They can work interactively to assess the effects of different characterizations of bias in a given word embedding or language model, which helps to specify informal intuitions in concrete resources for systematic testing.

With NLP research now quickly being transferred into real-world applications, it is important to be aware of and think through the consequences of our scientific investigation. Such ethical considerations are important in both authoring and reviewing. This tutorial will equip participants with basic guidelines for thinking deeply about ethical issues and review common considerations that recur in NLP research. The methodology is interactive and participatory, including case studies and working in groups. Importantly, the participants will be co-building the tutorial outcomes and will be working to create further tutorial materials to share as public outcomes.

In this paper, we present the results of a manual classification of all ethical consideration sections for ACL 2021. We also compare how many papers had an ethics consideration section per track and per world region in ACL 2021. We classified papers according to the ethical issues covered (research benefits, potential harms, and vulnerable groups affected) and whether the paper was marked as requiring ethics review by at least one reviewer. Moreover, we discuss recurring obstacles we have observed (highlighting some interesting texts we found along the way) and conclude with three suggestions. We think that this paper may be useful for anyone who needs to write — or review — an ethics section and would like to get an overview of what others have done.

Referring resolution is the task of identifying the referent of a natural language expression, for example “the woman behind the other woman getting a massage”. In this paper we investigate which are the kinds of referring expressions on which current transformer based models fail. Motivated by this analysis we identify the weakening of the spatial natural constraints as one of its causes and propose a model that aims to restore it. We evaluate our proposed model on different datasets for the task showing improved performance on the most challenging kinds of referring expressions. Finally we present a thorough analysis of the kinds errors that are improved by the new model and those that are not and remain future challenges for the task.

In order to interpret the communicative intents of an utterance, it needs to be grounded in something that is outside of language; that is, grounded in world modalities. In this paper, we argue that dialogue clarification mechanisms make explicit the process of interpreting the communicative intents of the speaker’s utterances by grounding them in the various modalities in which the dialogue is situated. This paper frames dialogue clarification mechanisms as an understudied research problem and a key missing piece in the giant jigsaw puzzle of natural language understanding. We discuss both the theoretical background and practical challenges posed by this problem and propose a recipe for obtaining grounding annotations. We conclude by highlighting ethical issues that need to be addressed in future work.

In this paper, we define and evaluate a methodology for extracting history-dependent spatial questions from visual dialogues. We say that a question is history-dependent if it requires (parts of) its dialogue history to be interpreted. We argue that some kinds of visual questions define a context upon which a follow-up spatial question relies. We call the question that restricts the context: trigger, and we call the spatial question that requires the trigger question to be answered: zoomer. We automatically extract different trigger and zoomer pairs based on the visual property that the questions rely on (e.g. color, number). We manually annotate the automatically extracted trigger and zoomer pairs to verify which zoomers require their trigger. We implement a simple baseline architecture based on a SOTA multimodal encoder. Our results reveal that there is much room for improvement for answering history-dependent questions.

In the visual dialog task GuessWhat?! two players maintain a dialog in order to identify a secret object in an image. Computationally, this is modeled using a question generation module and a guesser module for the questioner role and an answering model, the Oracle, to answer the generated questions. This raises a question: what’s the risk of having an imperfect oracle model?. Here we present work in progress in the study of the impact of different answering models in human generated questions in GuessWhat?!. We show that having access to better quality answers has a direct impact on the guessing task for human dialog and argue that better answers could help train better question generation models.

Collaborative grounding is a fundamental aspect of human-human dialog which allows people to negotiate meaning. In this paper we argue that it is missing from current deep learning approaches to dialog. Our central point is that making mistakes and being able to recover from them collaboratively is a key ingredient in grounding meaning. We illustrate the pitfalls of being unable to ground collaboratively, discuss what can be learned from the language acquisition and dialog systems literature, and reflect on how to move forward.

Visual Dialog is assumed to require the dialog history to generate correct responses during a dialog. However, it is not clear from previous work how dialog history is needed for visual dialog. In this paper we define what it means for a visual question to require dialog history and we release a subset of the Guesswhat?! questions for which their dialog history completely changes their responses. We propose a novel interpretable representation that visually grounds dialog history: the Region under Discussion. It constrains the image’s spatial features according to a semantic representation of the history inspired by the information structure notion of Question under Discussion.We evaluate the architecture on task-specific multimodal models and the visual transformer model LXMERT.

An interesting challenge for situated dialogue systems is referential visual dialog: by asking questions, the system has to identify the referent to which the user refers to. Task success is the standard metric used to evaluate these systems. However, it does not consider how effective each question is, that is how much each question contributes to the goal. We propose a new metric, that measures question effectiveness. As a preliminary study, we report the new metric for state of the art publicly available models on GuessWhat?!. Surprisingly, successful dialogues do not have a higher percentage of effective questions than failed dialogues. This suggests that a system with high task success is not necessarily one that generates good questions.

Task success is the standard metric used to evaluate referential visual dialogue systems. In this paper we propose two new metrics that evaluate how each question contributes to the goal. First, we measure how effective each question is by evaluating whether the question discards objects that are not the referent. Second, we define referring questions as those that univocally identify one object in the image. We report the new metrics for human dialogues and for state of the art publicly available models on GuessWhat?!. Regarding our first metric, we find that successful dialogues do not have a higher percentage of effective questions for most models. With respect to the second metric, humans make questions at the end of the dialogue that are referring, confirming their guess before guessing. Human dialogues that use this strategy have a higher task success but models do not seem to learn it.

In this paper, we study the grounding skills required to answer spatial questions asked by humans while playing the GuessWhat?! game. We propose a classification for spatial questions dividing them into absolute, relational, and group questions. We build a new answerer model based on the LXMERT multimodal transformer and we compare a baseline with and without visual features of the scene. We are interested in studying how the attention mechanisms of LXMERT are used to answer spatial questions since they require putting attention on more than one region simultaneously and spotting the relation holding among them. We show that our proposed model outperforms the baseline by a large extent (9.70% on spatial questions and 6.27% overall). By analyzing LXMERT errors and its attention mechanisms, we find that our classification helps to gain a better understanding of the skills required to answer different spatial questions.

In this paper we investigate the task of modeling how long it would take a student to respond to a tutor question during a tutoring dialogue. Solving such a task has applications in educational settings such as intelligent tutoring systems, as well as in platforms that help busy human tutors to keep students engaged. Knowing how long it would normally take a student to respond to different types of questions could help tutors optimize their own time while answering multiple dialogues concurrently, as well as deciding when to prompt a student again. We study this problem using data from a service that offers tutor support for math, chemistry and physics through an instant messaging platform. We create a dataset of 240K questions. We explore several strong baselines for this task and compare them with human performance.