Lambert Mathias

Rabeeh Karimi Mahabadi, Luke Zettlemoyer, James Henderson et al. · meta-ai, uw

Current methods for few-shot fine-tuning of pretrained masked language models (PLMs) require carefully engineered prompts and verbalizers for each new task to convert examples into a cloze-format that the PLM can score. In this work, we propose PERFECT, a simple and efficient method for few-shot fine-tuning of PLMs without relying on any such handcrafting, which is highly effective given as few as 32 data points. PERFECT makes two key design choices: First, we show that manually engineered task prompts can be replaced with task-specific adapters that enable sample-efficient fine-tuning and reduce memory and storage costs by roughly factors of 5 and 100, respectively. Second, instead of using handcrafted verbalizers, we learn new multi-token label embeddings during fine-tuning, which are not tied to the model vocabulary and which allow us to avoid complex auto-regressive decoding. These embeddings are not only learnable from limited data but also enable nearly 100x faster training and inference. Experiments on a wide range of few-shot NLP tasks demonstrate that PERFECT, while being simple and efficient, also outperforms existing state-of-the-art few-shot learning methods. Our code is publicly available at https://github.com/facebookresearch/perfect.git.

Badr AlKhamissi, Faisal Ladhak, Srini Iyer et al. · berkeley, meta-ai

Hate speech detection is complex; it relies on commonsense reasoning, knowledge of stereotypes, and an understanding of social nuance that differs from one culture to the next. It is also difficult to collect a large-scale hate speech annotated dataset. In this work, we frame this problem as a few-shot learning task, and show significant gains with decomposing the task into its "constituent" parts. In addition, we see that infusing knowledge from reasoning datasets (e.g. Atomic2020) improves the performance even further. Moreover, we observe that the trained models generalize to out-of-distribution datasets, showing the superiority of task decomposition and knowledge infusion compared to previously used methods. Concretely, our method outperforms the baseline by 17.83% absolute gain in the 16-shot case.

Suzanna Sia, Anton Belyy, Amjad Almahairi et al. · uw

Evaluating an explanation's faithfulness is desired for many reasons such as trust, interpretability and diagnosing the sources of model's errors. In this work, which focuses on the NLI task, we introduce the methodology of Faithfulness-through-Counterfactuals, which first generates a counterfactual hypothesis based on the logical predicates expressed in the explanation, and then evaluates if the model's prediction on the counterfactual is consistent with that expressed logic (i.e. if the new formula is \textit{logically satisfiable}). In contrast to existing approaches, this does not require any explanations for training a separate verification model. We first validate the efficacy of automatic counterfactual hypothesis generation, leveraging on the few-shot priming paradigm. Next, we show that our proposed metric distinguishes between human-model agreement and disagreement on new counterfactual input. In addition, we conduct a sensitivity analysis to validate that our metric is sensitive to unfaithful explanations.

Neema Kotonya, Andreas Vlachos, Majid Yazdani et al. · cambridge, meta-ai

Policy Compliance Detection (PCD) is a task we encounter when reasoning over texts, e.g. legal frameworks. Previous work to address PCD relies heavily on modeling the task as a special case of Recognizing Textual Entailment. Entailment is applicable to the problem of PCD, however viewing the policy as a single proposition, as opposed to multiple interlinked propositions, yields poor performance and lacks explainability. To address this challenge, more recent proposals for PCD have argued for decomposing policies into expression trees consisting of questions connected with logic operators. Question answering is used to obtain answers to these questions with respect to a scenario. Finally, the expression tree is evaluated in order to arrive at an overall solution. However, this work assumes expression trees are provided by experts, thus limiting its applicability to new policies. In this work, we learn how to infer expression trees automatically from policy texts. We ensure the validity of the inferred trees by introducing constrained decoding using a finite state automaton to ensure the generation of valid trees. We determine through automatic evaluation that 63% of the expression trees generated by our constrained generation model are logically equivalent to gold trees. Human evaluation shows that 88% of trees generated by our model are correct.

Wang-Chiew Tan, Jane Dwivedi-Yu, Yuliang Li et al.

Lifelogs are descriptions of experiences that a person had during their life. Lifelogs are created by fusing data from the multitude of digital services, such as online photos, maps, shopping and content streaming services. Question answering over lifelogs can offer personal assistants a critical resource when they try to provide advice in context. However, obtaining answers to questions over lifelogs is beyond the current state of the art of question answering techniques for a variety of reasons, the most pronounced of which is that lifelogs combine free text with some degree of structure such as temporal and geographical information. We create and publicly release TimelineQA1, a benchmark for accelerating progress on querying lifelogs. TimelineQA generates lifelogs of imaginary people. The episodes in the lifelog range from major life episodes such as high school graduation to those that occur on a daily basis such as going for a run. We describe a set of experiments on TimelineQA with several state-of-the-art QA models. Our experiments reveal that for atomic queries, an extractive QA system significantly out-performs a state-of-the-art retrieval-augmented QA system. For multi-hop queries involving aggregates, we show that the best result is obtained with a state-of-the-art table QA technique, assuming the ground truth set of episodes for deriving the answer is available.

Aaron Mueller, Kanika Narang, Lambert Mathias et al. · meta-ai

Large language models show impressive results on few-shot NLP tasks. However, these models are memory and computation-intensive. Meta-training allows one to leverage smaller models for few-shot generalization in a domain-general and task-agnostic manner; however, these methods alone results in models that may not have sufficient parameterization or knowledge to adapt quickly to a large variety of tasks. To overcome this issue, we propose meta-training with demonstration retrieval, where we use a dense passage retriever to retrieve semantically similar labeled demonstrations to each example for more varied supervision. By separating external knowledge from model parameters, we can use meta-training to train parameter-efficient models that generalize well on a larger variety of tasks. We construct a meta-training set from UnifiedQA and CrossFit, and propose a demonstration bank based on UnifiedQA tasks. To our knowledge, our work is the first to combine retrieval with meta-training, to use DPR models to retrieve demonstrations, and to leverage demonstrations from many tasks simultaneously, rather than randomly sampling demonstrations from the training set of the target task. Our approach outperforms a variety of targeted parameter-efficient and retrieval-augmented few-shot methods on QA, NLI, and text classification tasks (including SQuAD, QNLI, and TREC). Our approach can be meta-trained and fine-tuned quickly on a single GPU.

Charig Yang, Samiul Alam, Shakhrul Iman Siam et al.

To enable egocentric contextual AI in always-on smart glasses, it is crucial to be able to keep a record of the user's interactions with the world, including during reading. In this paper, we introduce a new task of reading recognition to determine when the user is reading. We first introduce the first-of-its-kind large-scale multimodal Reading in the Wild dataset, containing 100 hours of reading and non-reading videos in diverse and realistic scenarios. We then identify three modalities (egocentric RGB, eye gaze, head pose) that can be used to solve the task, and present a flexible transformer model that performs the task using these modalities, either individually or combined. We show that these modalities are relevant and complementary to the task, and investigate how to efficiently and effectively encode each modality. Additionally, we show the usefulness of this dataset towards classifying types of reading, extending current reading understanding studies conducted in constrained settings to larger scale, diversity and realism.

Jeonghwan Kim, Renjie Tao, Sanat Sharma et al.

Visual Question Answering (VQA) often requires coupling fine-grained perception with factual knowledge beyond the input image. Prior multimodal Retrieval-Augmented Generation (MM-RAG) systems improve factual grounding but lack an internal policy for when and how to retrieve. We propose PixSearch, the first end-to-end Segmenting Large Multimodal Model (LMM) that unifies region-level perception and retrieval-augmented reasoning. During encoding, PixSearch emits <search> tokens to trigger retrieval, selects query modalities (text, image, or region), and generates pixel-level masks that directly serve as visual queries, eliminating the reliance on modular pipelines (detectors, segmenters, captioners, etc.). A two-stage supervised fine-tuning regimen with search-interleaved supervision teaches retrieval timing and query selection while preserving segmentation ability. On egocentric and entity-centric VQA benchmarks, PixSearch substantially improves factual consistency and generalization, yielding a 19.7% relative gain in accuracy on CRAG-MM compared to whole image retrieval, while retaining competitive reasoning performance on various VQA and text-only QA tasks.

Shima Imani, Seungwhan Moon, Lambert Mathias et al.

Reliable mathematical and scientific reasoning remains an open challenge for large vision-language models. Standard final-answer evaluation often masks reasoning errors, allowing silent failures to persist. To address this gap, we introduce TRACE, a framework for Transparent Reasoning And Consistency Evaluation that diagnoses reasoning trajectories rather than only end results. At its core, TRACE leverages Auxiliary Reasoning Sets, compact sub question answer pairs that decompose complex problems, evaluate intermediate steps through consistency-based metrics, and expose failures overlooked by standard evaluation. Our experiments show that consistency across ARS correlates with final-answer correctness and helps pinpoint the reasoning steps where failures arise, offering actionable signals for model improvement. Furthermore, TRACE defines confidence regions that distinguish reliable from unreliable reasoning paths, supporting effective filtering, debugging, and model refinement.

Aaron Chan, Maziar Sanjabi, Lambert Mathias et al.

An extractive rationale explains a language model's (LM's) prediction on a given task instance by highlighting the text inputs that most influenced the prediction. Ideally, rationale extraction should be faithful (reflective of LM's actual behavior) and plausible (convincing to humans), without compromising the LM's (i.e., task model's) task performance. Although attribution algorithms and select-predict pipelines are commonly used in rationale extraction, they both rely on certain heuristics that hinder them from satisfying all three desiderata. In light of this, we propose UNIREX, a flexible learning framework that generalizes rationale extractor optimization as follows: (1) specify architecture for a learned rationale extractor; (2) select explainability objectives (i.e., faithfulness and plausibility criteria); and (3) jointly the train task model and rationale extractor on the task using the selected objectives. UNIREX enables replacing prior works' heuristic design choices with a generic learned rationale extractor in (1) and optimizing it for all three desiderata in (2)-(3). To facilitate comparison between methods with respect to multiple desiderata, we introduce the Normalized Relative Gain (NRG) metric. Across five text classification datasets, our best UNIREX configuration outperforms baselines by an average of 32.9% NRG. Plus, we find that UNIREX-trained rationale extractors can even generalize to unseen datasets and tasks.

Yuning Mao, Lambert Mathias, Rui Hou et al.

Recent parameter-efficient language model tuning (PELT) methods manage to match the performance of fine-tuning with much fewer trainable parameters and perform especially well when training data is limited. However, different PELT methods may perform rather differently on the same task, making it nontrivial to select the most appropriate method for a specific task, especially considering the fast-growing number of new PELT methods and tasks. In light of model diversity and the difficulty of model selection, we propose a unified framework, UniPELT, which incorporates different PELT methods as submodules and learns to activate the ones that best suit the current data or task setup via gating mechanism. On the GLUE benchmark, UniPELT consistently achieves 1~4% gains compared to the best individual PELT method that it incorporates and even outperforms fine-tuning under different setups. Moreover, UniPELT generally surpasses the upper bound that takes the best performance of all its submodules used individually on each task, indicating that a mixture of multiple PELT methods may be inherently more effective than single methods.

Alireza Roshan-Ghias, Clint Solomon Mathialagan, Pragaash Ponnusamy et al.

Spoken language understanding (SLU) systems in conversational AI agents often experience errors in the form of misrecognitions by automatic speech recognition (ASR) or semantic gaps in natural language understanding (NLU). These errors easily translate to user frustrations, particularly so in recurrent events e.g. regularly toggling an appliance, calling a frequent contact, etc. In this work, we propose a query rewriting approach by leveraging users' historically successful interactions as a form of memory. We present a neural retrieval model and a pointer-generator network with hierarchical attention and show that they perform significantly better at the query rewriting task with the aforementioned user memories than without. We also highlight how our approach with the proposed models leverages the structural and semantic diversity in ASR's output towards recovering users' intents.

Zheng Chen, Xing Fan, Yuan Ling et al.

Query rewriting (QR) is an increasingly important technique to reduce customer friction caused by errors in a spoken language understanding pipeline, where the errors originate from various sources such as speech recognition errors, language understanding errors or entity resolution errors. In this work, we first propose a neural-retrieval based approach for query rewriting. Then, inspired by the wide success of pre-trained contextual language embeddings, and also as a way to compensate for insufficient QR training data, we propose a language-modeling (LM) based approach to pre-train query embeddings on historical user conversation data with a voice assistant. In addition, we propose to use the NLU hypotheses generated by the language understanding system to augment the pre-training. Our experiments show pre-training provides rich prior information and help the QR task achieve strong performance. We also show joint pre-training with NLU hypotheses has further benefit. Finally, after pre-training, we find a small set of rewrite pairs is enough to fine-tune the QR model to outperform a strong baseline by full training on all QR training data.

Adrian de Wynter, Lambert Mathias

Dealing with previously unseen slots is a challenging problem in a real-world multi-domain dialogue state tracking task. Other approaches rely on predefined mappings to generate candidate slot keys, as well as their associated values. This, however, may fail when the key, the value, or both, are not seen during training. To address this problem we introduce a neural network that leverages external knowledge bases (KBs) to better classify out-of-vocabulary slot keys and values. This network projects the slot into an attribute space derived from the KB, and, by leveraging similarities in this space, we propose candidate slot keys and values to the dialogue state tracker. We provide extensive experiments that demonstrate that our stratagem can improve upon a previous approach, which relies on predefined candidate mappings. In particular, we evaluate this approach by training a state-of-the-art model with candidates generated from our network, and obtained relative increases of 57.7% and 82.7% in F1 score and accuracy, respectively, for the aforementioned model, when compared to the current candidate generation strategy.

Rylan Conway, Lambert Mathias

In a spoken dialogue system, dialogue state tracker (DST) components track the state of the conversation by updating a distribution of values associated with each of the slots being tracked for the current user turn, using the interactions until then. Much of the previous work has relied on modeling the natural order of the conversation, using distance based offsets as an approximation of time. In this work, we hypothesize that leveraging the wall-clock temporal difference between turns is crucial for finer-grained control of dialogue scenarios. We develop a novel approach that applies a {\it time mask}, based on the wall-clock time difference, to the associated slot embeddings and empirically demonstrate that our proposed approach outperforms existing approaches that leverage distance offsets, on both an internal benchmark dataset as well as DSTC2.

Tongfei Chen, Chetan Naik, Hua He et al.

Tracking the state of the conversation is a central component in task-oriented spoken dialogue systems. One such approach for tracking the dialogue state is slot carryover, where a model makes a binary decision if a slot from the context is relevant to the current turn. Previous work on the slot carryover task used models that made independent decisions for each slot. A close analysis of the results show that this approach results in poor performance over longer context dialogues. In this paper, we propose to jointly model the slots. We propose two neural network architectures, one based on pointer networks that incorporate slot ordering information, and the other based on transformer networks that uses self attention mechanism to model the slot interdependencies. Our experiments on an internal dialogue benchmark dataset and on the public DSTC2 dataset demonstrate that our proposed models are able to resolve longer distance slot references and are able to achieve competitive performance.

Michael Regan, Pushpendre Rastogi, Arpit Gupta et al.

We present Contextual Query Rewrite (CQR) a dataset for multi-domain task-oriented spoken dialogue systems that is an extension of the Stanford dialog corpus (Eric et al., 2017a). While previous approaches have addressed the issue of diverse schemas by learning candidate transformations (Naik et al., 2018), we instead model the reference resolution task as a user query reformulation task, where the dialog state is serialized into a natural language query that can be executed by the downstream spoken language understanding system. In this paper, we describe our methodology for creating the query reformulation extension to the dialog corpus, and present an initial set of experiments to establish a baseline for the CQR task. We have released the corpus to the public [1] to support further research in this area.

Pushpendre Rastogi, Arpit Gupta, Tongfei Chen et al.

We present a novel approach to dialogue state tracking and referring expression resolution tasks. Successful contextual understanding of multi-turn spoken dialogues requires resolving referring expressions across turns and tracking the entities relevant to the conversation across turns. Tracking conversational state is particularly challenging in a multi-domain scenario when there exist multiple spoken language understanding (SLU) sub-systems, and each SLU sub-system operates on its domain-specific meaning representation. While previous approaches have addressed the disparate schema issue by learning candidate transformations of the meaning representation, in this paper, we instead model the reference resolution as a dialogue context-aware user query reformulation task -- the dialog state is serialized to a sequence of natural language tokens representing the conversation. We develop our model for query reformulation using a pointer-generator network and a novel multi-task learning setup. In our experiments, we show a significant improvement in absolute F1 on an internal as well as a, soon to be released, public benchmark respectively.

Amr Sharaf, Arpit Gupta, Hancheng Ge et al.

In the slot-filling paradigm, where a user can refer back to slots in the context during the conversation, the goal of the contextual understanding system is to resolve the referring expressions to the appropriate slots in the context. In this paper, we build on the context carryover system~\citep{Naik2018ContextualSC}, which provides a scalable multi-domain framework for resolving references. However, scaling this approach across languages is not a trivial task, due to the large demand on acquisition of annotated data in the target language. Our main focus is on cross-lingual methods for reference resolution as a way to alleviate the need for annotated data in the target language. In the cross-lingual setup, we assume there is access to annotated resources as well as a well trained model in the source language and little to no annotated data in the target language. In this paper, we explore three different approaches for cross-lingual transfer \textemdash~\ delexicalization as data augmentation, multilingual embeddings and machine translation. We compare these approaches both on a low resource setting as well as a large resource setting. Our experiments show that multilingual embeddings and delexicalization via data augmentation have a significant impact in the low resource setting, but the gains diminish as the amount of available data in the target language increases. Furthermore, when combined with machine translation we can get performance very close to actual live data in the target language, with only 25\% of the data projected into the target language.

Chetan Naik, Arpit Gupta, Hancheng Ge et al.

In the slot-filling paradigm, where a user can refer back to slots in the context during a conversation, the goal of the contextual understanding system is to resolve the referring expressions to the appropriate slots in the context. In large-scale multi-domain systems, this presents two challenges - scaling to a very large and potentially unbounded set of slot values, and dealing with diverse schemas. We present a neural network architecture that addresses the slot value scalability challenge by reformulating the contextual interpretation as a decision to carryover a slot from a set of possible candidates. To deal with heterogenous schemas, we introduce a simple data-driven method for trans- forming the candidate slots. Our experiments show that our approach can scale to multiple domains and provides competitive results over a strong baseline.

Xing Fan, Emilio Monti, Lambert Mathias et al.

The goal of semantic parsing is to map natural language to a machine interpretable meaning representation language (MRL). One of the constraints that limits full exploration of deep learning technologies for semantic parsing is the lack of sufficient annotation training data. In this paper, we propose using sequence-to-sequence in a multi-task setup for semantic parsing with a focus on transfer learning. We explore three multi-task architectures for sequence-to-sequence modeling and compare their performance with an independently trained model. Our experiments show that the multi-task setup aids transfer learning from an auxiliary task with large labeled data to a target task with smaller labeled data. We see absolute accuracy gains ranging from 1.0% to 4.4% in our in- house data set, and we also see good gains ranging from 2.5% to 7.0% on the ATIS semantic parsing tasks with syntactic and semantic auxiliary tasks.