LLMs instead of Human Judges? A Large Scale Empirical Study across 20 NLP Evaluation Tasks

There is an increasing trend towards evaluating NLP models with LLM-generated judgments instead of human judgments. In the absence of a comparison against human data, this raises concerns about the validity of these evaluations; in case they are conducted with proprietary models, this also raises concerns over reproducibility. We provide JUDGE-BENCH, a collection of 20 NLP datasets with human annotations, and comprehensively evaluate 11 current LLMs, covering both open-weight and proprietary models, for their ability to replicate the annotations. Our evaluations show that each LLM exhibits a large variance across datasets in its correlation to human judgments. We conclude that LLMs are not yet ready to systematically replace human judges in NLP.

Learning to Ask Informative Questions: Enhancing LLMs with Preference Optimization and Expected Information Gain

Questions are essential tools for acquiring the necessary information to complete information-seeking tasks. However, large language models (LLMs), especially open-source models, often perform poorly in generating informative questions, as measured by expected information gain (EIG). In this paper, we propose a method to enhance the informativeness of LLM-generated questions in 20-question game dialogues. We sample multiple questions from the same model (LLAMA 2-CHAT 7B) for each game and create pairs of low-EIG and high-EIG questions to apply a Direct Preference Optimization (DPO) algorithm. Our results show that this method produces more effective questions (in terms of EIG), even in domains different from those used to train the DPO model.

A Systematic Analysis of Large Language Models as Soft Reasoners: The Case of Syllogistic Inferences

The reasoning abilities of Large Language Models (LLMs) are becoming a central focus of study in NLP. In this paper, we consider the case of syllogistic reasoning, an area of deductive reasoning studied extensively in logic and cognitive psychology. Previous research has shown that pre-trained LLMs exhibit reasoning biases, such as $\textit{content effects}$, avoid answering that $\textit{no conclusion follows}$, display human-like difficulties, and struggle with multi-step reasoning. We contribute to this research line by systematically investigating the effects of chain-of-thought reasoning, in-context learning (ICL), and supervised fine-tuning (SFT) on syllogistic reasoning, considering syllogisms with conclusions that support or violate world knowledge, as well as ones with multiple premises. Crucially, we go beyond the standard focus on accuracy, with an in-depth analysis of the conclusions generated by the models. Our results suggest that the behavior of pre-trained LLMs can be explained by heuristics studied in cognitive science and that both ICL and SFT improve model performance on valid inferences, although only the latter mitigates most reasoning biases without harming model consistency.

Looking for Confirmations: An Effective and Human-Like Visual Dialogue Strategy

Generating goal-oriented questions in Visual Dialogue tasks is a challenging and long-standing problem. State-Of-The-Art systems are shown to generate questions that, although grammatically correct, often lack an effective strategy and sound unnatural to humans. Inspired by the cognitive literature on information search and cross-situational word learning, we design Confirm-it, a model based on a beam search re-ranking algorithm that guides an effective goal-oriented strategy by asking questions that confirm the model's conjecture about the referent. We take the GuessWhat?! game as a case-study. We show that dialogues generated by Confirm-it are more natural and effective than beam search decoding without re-ranking.

Overprotective Training Environments Fall Short at Testing Time: Let Models Contribute to Their Own Training

Despite important progress, conversational systems often generate dialogues that sound unnatural to humans. We conjecture that the reason lies in their different training and testing conditions: agents are trained in a controlled "lab" setting but tested in the "wild". During training, they learn to generate an utterance given the human dialogue history. On the other hand, during testing, they must interact with each other, and hence deal with noisy data. We propose to fill this gap by training the model with mixed batches containing both samples of human and machine-generated dialogues. We assess the validity of the proposed method on GuessWhat?!, a visual referential game.

The Interplay of Task Success and Dialogue Quality: An in-depth Evaluation in Task-Oriented Visual Dialogues

When training a model on referential dialogue guessing games, the best model is usually chosen based on its task success. We show that in the popular end-to-end approach, this choice prevents the model from learning to generate linguistically richer dialogues, since the acquisition of language proficiency takes longer than learning the guessing task. By comparing models playing different games (GuessWhat, GuessWhich, and Mutual Friends), we show that this discrepancy is model- and task-agnostic. We investigate whether and when better language quality could lead to higher task success. We show that in GuessWhat, models could increase their accuracy if they learn to ground, encode, and decode also words that do not occur frequently in the training set.

Psycholinguistics meets Continual Learning: Measuring Catastrophic Forgetting in Visual Question Answering

We study the issue of catastrophic forgetting in the context of neural multimodal approaches to Visual Question Answering (VQA). Motivated by evidence from psycholinguistics, we devise a set of linguistically-informed VQA tasks, which differ by the types of questions involved (Wh-questions and polar questions). We test what impact task difficulty has on continual learning, and whether the order in which a child acquires question types facilitates computational models. Our results show that dramatic forgetting is at play and that task difficulty and order matter. Two well-known current continual learning methods mitigate the problem only to a limiting degree.

Evaluating the Representational Hub of Language and Vision Models

The multimodal models used in the emerging field at the intersection of computational linguistics and computer vision implement the bottom-up processing of the `Hub and Spoke' architecture proposed in cognitive science to represent how the brain processes and combines multi-sensory inputs. In particular, the Hub is implemented as a neural network encoder. We investigate the effect on this encoder of various vision-and-language tasks proposed in the literature: visual question answering, visual reference resolution, and visually grounded dialogue. To measure the quality of the representations learned by the encoder, we use two kinds of analyses. First, we evaluate the encoder pre-trained on the different vision-and-language tasks on an existing diagnostic task designed to assess multimodal semantic understanding. Second, we carry out a battery of analyses aimed at studying how the encoder merges and exploits the two modalities.

Jointly Learning to See, Ask, and GuessWhat

We are interested in understanding how the ability to ground language in vision interacts with other abilities at play in dialogue, such as asking a series of questions to obtain the necessary information to perform a certain task. With this aim, we develop a Questioner agent in the context of the GuessWhat?! game. Our model exploits a neural network architecture to build a continuous representation of the dialogue state that integrates information from the visual and linguistic modalities and conditions future action. To play the GuessWhat?! game, the Questioner agent has to be able to do both, ask questions and guess a target object in the visual environment. In our architecture, these two capabilities are considered jointly as a supervised multi-task learning problem, to which cooperative learning can be further applied. We show that the introduction of our new architecture combined with these learning regimes yields an increase of 19.5% in task success accuracy with respect to a baseline model that treats submodules independently. With this increase, we reach an accuracy comparable to state-of-the-art models that use reinforcement learning, with the advantage that our architecture is entirely differentiable and thus easier to train. This suggests that combining our approach with reinforcement learning could lead to further improvements in the future. Finally, we present a range of analyses that examine the quality of the dialogues and shed light on the internal dynamics of the model.

Grounded Textual Entailment

Capturing semantic relations between sentences, such as entailment, is a long-standing challenge for computational semantics. Logic-based models analyse entailment in terms of possible worlds (interpretations, or situations) where a premise P entails a hypothesis H iff in all worlds where P is true, H is also true. Statistical models view this relationship probabilistically, addressing it in terms of whether a human would likely infer H from P. In this paper, we wish to bridge these two perspectives, by arguing for a visually-grounded version of the Textual Entailment task. Specifically, we ask whether models can perform better if, in addition to P and H, there is also an image (corresponding to the relevant "world" or "situation"). We use a multimodal version of the SNLI dataset (Bowman et al., 2015) and we compare "blind" and visually-augmented models of textual entailment. We show that visual information is beneficial, but we also conduct an in-depth error analysis that reveals that current multimodal models are not performing "grounding" in an optimal fashion.