More Like this

1. Bidirectional Language Models Are Also Few-shot Learners

   Patel, Ajay; Li, Bryan; Rasooli, Mohammad Sadegh; Constant, Noah; Raffel, Colin; Callison-Burch, Chris (May 2023, The Eleventh International Conference on Learning Representations (ICLR 2023))

   Large language models such as GPT-3 (Brown et al., 2020) can perform arbitrary tasks without undergoing fine-tuning after being prompted with only a few labeled examples. An arbitrary task can be reformulated as a natural language prompt, and a language model can be asked to generate the completion, indirectly performing the task in a paradigm known as prompt-based learning. To date, emergent prompt-based learning capabilities have mainly been demonstrated for unidirectional language models. However, bidirectional language models pre-trained on denoising objectives such as masked language modeling produce stronger learned representations for transfer learning. This motivates the possibility of prompting bidirectional models, but their pre-training objectives have made them largely incompatible with the existing prompting paradigm. We present SAP (Sequential Autoregressive Prompting), a technique that enables the prompting of bidirectional models. Utilizing the machine translation task as a case study, we prompt the bidirectional mT5 model (Xue et al., 2021) with SAP and demonstrate its few-shot and zero-shot translations outperform the few-shot translations of unidirectional models like GPT-3 and XGLM (Lin et al., 2021), despite mT5's approximately 50% fewer parameters. We further show SAP is effective on question answering and summarization. For the first time, our results demonstrate prompt-based learning is an emergent property of a broader class of language models, rather than only unidirectional models. 

   more » « less
2. Task Descriptors Help Transformers Learn Linear Models In-Context

   Huang, Ruomin; Ge, Rong (April 2025, ICLR 2025)

   Large language models (LLMs) exhibit strong in-context learning (ICL) ability, which allows the model to make predictions on new examples based on the given prompt. Recently, a line of research (Von Oswald et al., 2023; Aky¨urek et al., 2023; Ahn et al., 2023; Mahankali et al., 2023; Zhang et al., 2024) considered ICL for a simple linear regression setting and showed that the forward pass of Transformers is simulating some variants of gradient descent (GD) algorithms on the in-context examples. In practice, the input prompt usually contains a task descriptor in addition to in-context examples. We investigate how the task description helps ICL in the linear regression setting. Consider a simple setting where the task descriptor describes the mean of input in linear regression. Our results show that gradient flow converges to a global minimum for a linear Transformer. At the global minimum, the Transformer learns to use the task descriptor effectively to improve its performance. Empirically, we verify our results by showing that the weights converge to the predicted global minimum and Transformers indeed perform better with task descriptors. 

   more » « less
3. Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?

   https://doi.org/10.1162/tacl_a_00548  

   Oh, Byung-Doh; Schuler, William (January 2023, Transactions of the Association for Computational Linguistics)

   Abstract This work presents a linguistic analysis into why larger Transformer-based pre-trained language models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times. First, regression analyses show a strictly monotonic, positive log-linear relationship between perplexity and fit to reading times for the more recently released five GPT-Neo variants and eight OPT variants on two separate datasets, replicating earlier results limited to just GPT-2 (Oh et al., 2022). Subsequently, analysis of residual errors reveals a systematic deviation of the larger variants, such as underpredicting reading times of named entities and making compensatory overpredictions for reading times of function words such as modals and conjunctions. These results suggest that the propensity of larger Transformer-based models to ‘memorize’ sequences during training makes their surprisal estimates diverge from humanlike expectations, which warrants caution in using pre-trained language models to study human language processing. 

   more » « less
4. Work in Progress: Using Machine Learning to Map Student Narratives of Understanding and Promoting Linguistic Justice

   Auby, H.; Koretsky, M. (June 2023, 2023 ASEE Annual Conference & Exposition)

   his work-in-progress paper expands on a collaboration between engineering education researchers and machine learning researchers to automate the analysis of written responses to conceptually challenging questions in statics and dynamics courses (Authors, 2022). Using the Concept Warehouse (Koretsky et al., 2014), written justifications of ConcepTests (CTs) were gathered from statics and dynamics courses in a diverse set of two- and four-year institutions. Written justifications for CTs have been used to support active learning pedagogies which makes them important to investigate how students put together their problem-solving narratives of understanding. However, despite the large benefit that analysis of student written responses may provide to instructors and researchers, manual review of responses is cumbersome, limits analysis, and can be prone to human bias. In efforts to improve the analysis of student written responses, machine learning has been used in various educational contexts to analyze short and long texts (Burstein et al., 2020; Burstein et al., 2021). Natural Language Processing (NLP) uses transformer-based machine learning models (Brown et al., 2020; Raffel et al., 2019) which can be used through fine-tuning or in-context learning methods. NLP can be used to train algorithms that can automate the coding of written responses. Only a few studies for educational applications have leveraged transformer-based machine learning models further prompting an investigation into its use in STEM education. However, work in NLP has been criticized for heightening the possibility to perpetuate and even amplify harmful stereotypes and implicit biases (Chang et al., 2019; Mayfield et al., 2019). In this study, we detail the aim to use NLP for linguistic justice. Using methods like text summary, topic modeling, and text classification, we identify key aspects of student narratives of understanding in written responses to mechanics and statics CTs. Through this process, we seek to use machine learning to identify different ways students talk about a problem and their understanding at any point in their narrative formation process. Thus, we hope to help reduce human bias in the classroom and through technology by giving instructors and researchers a diverse set of narratives that include insight into their students’ histories, identities, and understanding. These can then be used towards connecting technological knowledge to students’ everyday lives. 

   more » « less
5. A Logic for Expressing Log-Precision Transformers

   Merrill, William; Sabharwal, Ashish (December 2023, NeurIPS 2023)

   One way to interpret the reasoning power of transformer-based language models is to describe the types of logical rules they can resolve over some input text. Recently, Chiang et al. (2023) showed that finite-precision transformers can be equivalently expressed in a generalization of first-order logic. However, finite-precision transformers are a weak transformer variant because, as we show, a single head can only attend to a constant number of tokens and, in particular, cannot represent uniform attention. Since attending broadly is a core capability for transformers, we ask whether a minimally more expressive model that can attend universally can also be characterized in logic. To this end, we analyze transformers whose forward pass is computed in logn precision on contexts of length n. We prove that any log-precision transformer can be equivalently expressed as a first-order logic sentence that, in addition to standard universal and existential quantifiers, may also contain majority-vote quantifiers. This is the tightest known upper bound and first logical characterization of log-precision transformers. 

   more » « less