Large language models (LLMs) offer impressive performance in various zero-shot and few-shot tasks. However, their success in zero-shot and few-shot settings may be affected by task contamination, a potential limitation that has not been thoroughly examined. This paper investigates how zero-shot and few-shot performance of LLMs has changed chronologically over time. Utilizing GPT-3 series models and several other recent open-sourced LLMs, and controlling for dataset difficulty, we find that on datasets released before the LLM training data creation date, LLMs perform surprisingly better than on datasets released after. This strongly indicates that, for many LLMs, there exists task contamination on zero-shot and few-shot evaluation for datasets released prior to the LLMs' training data creation date. Additionally, we utilize training data inspection, task example extraction, and a membership inference attack, which reveal further evidence of task contamination. Importantly, we find that for classification tasks with no possibility of task contamination, LLMs rarely demonstrate statistically significant improvements over simple majority baselines, in both zero and few-shot settings.
The phenomenon of model-wise double descent, where the test error peaks and then reduces as the model size increases, is an interesting topic that has attracted the attention of researchers due to the striking observed gap between theory and practice \citep{Belkin2018ReconcilingMM}. Additionally, while double descent has been observed in various tasks and architectures, the peak of double descent can sometimes be noticeably absent or diminished, even without explicit regularization, such as weight decay and early stopping. In this paper, we investigate this intriguing phenomenon from the optimization perspective and propose a simple optimization-based explanation for why double descent sometimes occurs weakly or not at all. To the best of our knowledge, we are the first to demonstrate that many disparate factors contributing to model-wise double descent (initialization, normalization, batch size, learning rate, optimization algorithm) are unified from the viewpoint of optimization: model-wise double descent is observed if and only if the optimizer can find a sufficiently low-loss minimum. These factors directly affect the condition number of the optimization problem or the optimizer and thus affect the final minimum found by the optimizer, reducing or increasing the height of the double descent peak. We conduct a series of controlled experiments on random feature models and two-layer neural networks under various optimization settings, demonstrating this optimization-based unified view. Our results suggest the following implication: Double descent is unlikely to be a problem for real-world machine learning setups. Additionally, our results help explain the gap between weak double descent peaks in practice and strong peaks observable in carefully designed setups.
Verifying mathematical proofs is difficult, but can be automated with the assistance of a computer. Autoformalization is the task of automatically translating natural language mathematics into a formal language that can be verified by a program. This is a challenging task, and especially for higher-level mathematics found in research papers. Research paper mathematics requires large amounts of background and context. In this paper, we propose an avenue towards tackling autoformalization for research-level mathematics, by breaking the task into easier and more approachable subtasks: unlinked formalization (formalization with unlinked definitions and theorems), entity linking (linking to the proper theorems and definitions), and finally adjusting types so it passes the type checker. In addition, we present arXiv2Formal, a benchmark dataset for unlinked formalization consisting of 50 theorems formalized for the Lean theorem prover sampled from papers on arXiv.org. We welcome any contributions from the community to future versions of this dataset.
Figurative language is commonplace in natural language, and while making communication memorable and creative, can be difficult to understand. In this work, we investigate the robustness of Question Answering (QA) models on figurative text. Yes/no questions, in particular, are a useful probe of figurative language understanding capabilities of large language models. We propose FigurativeQA, a set of 1000 yes/no questions with figurative and non-figurative contexts, extracted from the domains of restaurant and product reviews. We show that state-of-the-art BERT-based QA models exhibit an average performance drop of up to 15\% points when answering questions from figurative contexts, as compared to non-figurative ones. While models like GPT-3 and ChatGPT are better at handling figurative texts, we show that further performance gains can be achieved by automatically simplifying the figurative contexts into their non-figurative (literal) counterparts. We find that the best overall model is ChatGPT with chain-of-thought prompting to generate non-figurative contexts. Our work provides a promising direction for building more robust QA models with figurative language understanding capabilities.
There has been significant interest in zero and few-shot learning for dialogue state tracking (DST) due to the high cost of collecting and annotating task-oriented dialogues. Recent work has demonstrated that in-context learning requires very little data and zero parameter updates, and even outperforms trained methods in the few-shot setting (Hu et al. 2022). We propose RefPyDST, which advances the state of the art with three advancements to in-context learning for DST. First, we formulate DST as a Python programming task, explicitly modeling language coreference as variable reference in Python. Second, since in-context learning depends highly on the context examples, we propose a method to retrieve a diverse set of relevant examples to improve performance. Finally, we introduce a novel re-weighting method during decoding that takes into account probabilities of competing surface forms, and produces a more accurate dialogue state prediction. We evaluate our approach using MultiWOZ and achieve state-of-the-art multi-domain joint-goal accuracy in zero and few-shot settings.
In this paper, we introduce Dependency Dialogue Acts (DDA), a novel framework for capturing the structure of speaker-intentions in multi-party dialogues. DDA combines and adapts features from existing dialogue annotation frameworks, and emphasizes the multi-relational response structure of dialogues in addition to the dialogue acts and rhetorical relations. It represents the functional, discourse, and response structure in multi-party multi-threaded conversations. A few key features distinguish DDA from existing dialogue annotation frameworks such as SWBD-DAMSL and the ISO 24617-2 standard. First, DDA prioritizes the relational structure of the dialogue units and the dialog context, annotating both dialog acts and rhetorical relations as response relations to particular utterances. Second, DDA embraces overloading in dialogues, encouraging annotators to specify multiple response relations and dialog acts for each dialog unit. Lastly, DDA places an emphasis on adequately capturing how a speaker is using the full dialog context to plan and organize their speech. With these features, DDA is highly expressive and recall-oriented with regard to conversation dynamics between multiple speakers. In what follows, we present the DDA annotation framework and case studies annotating DDA structures in multi-party, multi-threaded conversations.
Code-switching, or switching between languages, occurs for many reasons and has important linguistic, sociological, and cultural implications. Multilingual speakers code-switch for a variety of purposes, such as expressing emotions, borrowing terms, making jokes, introducing a new topic, etc. The reason for code-switching may be quite useful for analysis, but is not readily apparent. To remedy this situation, we annotate a new dataset of motivations for code-switching in Spanish-English. We build the first system (to our knowledge) to automatically identify a wide range of motivations that speakers code-switch in everyday speech, achieving an accuracy of 75% across all motivations. Additionally, we show that the system can be adapted to new language pairs, achieving 66% accuracy on a new language pair (Hindi-English), demonstrating the cross-lingual applicability of our annotation scheme
Popular models such as Transformers and LSTMs use tokens as its unit of information. That is, each token is encoded into a vector representation, and those vectors are used directly in a computation. However, humans frequently consider spans of tokens (i.e., phrases) instead of their constituent tokens. In this paper we introduce Treeformer, an architecture inspired by the CKY algorithm and Transformer which learns a composition operator and pooling function in order to construct hierarchical encodings for phrases and sentences. Our extensive experiments demonstrate the benefits of incorporating a hierarchical structure into the Transformer, and show significant improvements compared to a baseline Transformer in machine translation, abstractive summarization, and various natural language understanding tasks.
Despite extensive research on parsing of English sentences into Abstraction Meaning Representation (AMR) graphs, which are compared to gold graphs via the Smatch metric, full-document parsing into a unified graph representation lacks well-defined representation and evaluation. Taking advantage of a super-sentential level of coreference annotation from previous work, we introduce a simple algorithm for deriving a unified graph representation, avoiding the pitfalls of information loss from over-merging and lack of coherence from under-merging. Next, we describe improvements to the Smatch metric to make it tractable for comparing document-level graphs, and use it to re-evaluate the best published document-level AMR parser. We also present a pipeline approach combining the top performing AMR parser and coreference resolution systems, providing a strong baseline for future research.