The ability to perform arithmetic tasks is a remarkable trait of human intelligence and might form a critical component of more complex reasoning tasks. In this work, we investigate if the surface form of a number has any influence on how sequence-to-sequence language models learn simple arithmetic tasks such as addition and subtraction across a wide range of values. We find that how a number is represented in its surface form has a strong influence on the model's accuracy. In particular, the model fails to learn addition of five-digit numbers when using subwords (e.g., "32"), and it struggles to learn with character-level representations (e.g., "3 2"). By introducing position tokens (e.g., "3 10e1 2"), the model learns to accurately add and subtract numbers up to 60 digits. We conclude that modern pretrained language models can easily learn arithmetic from very few examples, as long as we use the proper surface representation. This result bolsters evidence that subword tokenizers and positional encodings are components in current transformer designs that might need improvement. Moreover, we show that regardless of the number of parameters and training examples, models cannot learn addition rules that are independent of the length of the numbers seen during training. Code to reproduce our experiments is available at https://github.com/castorini/transformers-arithmetic
How Numerical Precision Affects Arithmetical Reasoning Capabilities of LLMs
State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University（通用人工智能国家重点实验室，智能科学与技术学院，北京大学）；School of EECS, Peking University（电子工程系，北京大学）；Huawei Noah’s Ark Lab（华为诺亚实验室）；Center for Machine Learning Research, Peking University（机器学习研究中心，北京大学）
The transformer-based pre-trained language models have been tremendously successful in most of the conventional NLP tasks. But they often struggle in those tasks where numerical understanding is required. Some possible reasons can be the tokenizers and pre-training objectives which are not specifically designed to learn and preserve numeracy. Here we investigate the ability of text-to-text transfer learning model (T5), which has outperformed its predecessors in the conventional NLP tasks, to learn numeracy. We consider four numeracy tasks: numeration, magnitude order prediction, finding minimu…
We analyze how large language models (LLMs) represent out-of-context words, investigating their reliance on the given context to capture their semantics. Our likelihood-guided text perturbations reveal a correlation between token likelihood and attention values in transformer-based language models. Extensive experiments reveal that unexpected tokens cause the model to attend less to the information coming from themselves to compute their representations, particularly at higher layers. These findings have valuable implications for assessing the robustness of LLMs in real-world scenarios. Fully …
University of Cambridge（剑桥大学）；Microsoft（微软）；Imperial College London（帝国理工学院）
Large Language Models (LLMs) have demonstrated impressive performance on a range of Natural Language Processing (NLP) tasks. Unfortunately, the immense amount of computations and memory accesses required for LLM training makes them prohibitively expensive in terms of hardware cost, and thus challenging to deploy in use cases such as on-device learning. In this paper, motivated by the observation that LLM training is memory-bound, we propose a novel dynamic quantization strategy, termed Dynamic Stashing Quantization (DSQ), that puts a special focus on reducing the memory operations, but also en…
Sun Yat-sen University（中山大学）；Fudan University（复旦大学）；Microsoft Research（微软研究院）；SINOVATION VENTURES（创新工场）
Analytical reasoning is an essential and challenging task that requires a system to analyze a scenario involving a set of particular circumstances and perform reasoning over it to make conclusions. In this paper, we study the challenge of analytical reasoning of text and introduce a new dataset consisting of questions from the Law School Admission Test from 1991 to 2016. We analyze what knowledge understanding and reasoning abilities are required to do well on this task. Furthermore, to address this reasoning challenge, we design two different baselines: (1) a Transformer-based method which le…
This paper outlines the use of Transformer networks trained to translate math word problems to equivalent arithmetic expressions in infix, prefix, and postfix notations. We compare results produced by many neural configurations and find that most configurations outperform previously reported approaches on three of four datasets with significant increases in accuracy of over 20 percentage points. The best neural approaches boost accuracy by 30% when compared to the previous state-of-the-art on some datasets.
Investigating Task Arithmetic for Zero-Shot Information Retrieval
Department of Informatics, Systems and Communication - DISCo, University of Milano-Bicocca（信息学、系统与通信系 - DISCo，米兰-比科卡大学）；DAUIN Dipartimento di Automatica e Informatica, Politecnico di Torino（DAUIN 自动控制与信息系，托斯卡纳理工学院）
University of Colorado Colorado Springs（科罗拉多大学科罗拉多斯普林斯分校）
Constructing accurate and automatic solvers of math word problems has proven to be quite challenging. Prior attempts using machine learning have been trained on corpora specific to math word problems to produce arithmetic expressions in infix notation before answer computation. We find that custom-built neural networks have struggled to generalize well. This paper outlines the use of Transformer networks trained to translate math word problems to equivalent arithmetic expressions in infix, prefix, and postfix notations. In addition to training directly on domain-specific corpora, we use an app…
University of California, Santa Barbara（加州大学圣巴巴拉分校）
The growth of social media has encouraged the written use of African American Vernacular English (AAVE), which has traditionally been used only in oral contexts. However, NLP models have historically been developed using dominant English varieties, such as Standard American English (SAE), due to text corpora availability. We investigate the performance of GPT-2 on AAVE text by creating a dataset of intent-equivalent parallel AAVE/SAE tweet pairs, thereby isolating syntactic structure and AAVE- or SAE-specific language for each pair. We evaluate each sample and its GPT-2 generated text with pre…
While large pretrained Transformer models have proven highly capable at tackling natural language tasks, handling long sequence inputs continues to be a significant challenge. One such task is long input summarization, where inputs are longer than the maximum input context of most pretrained models. Through an extensive set of experiments, we investigate what model architectural changes and pretraining paradigms can most efficiently adapt a pretrained Transformer for long input summarization. We find that a staggered, block-local Transformer with global encoder tokens strikes a good balance of…
University of Arizona（亚利桑那大学）；Microsoft Research（微软研究院）；Allen Institute for AI（艾伦人工智能研究所）
In this work, we explore techniques for augmenting interactive agents with information from symbolic modules, much like humans use tools like calculators and GPS systems to assist with arithmetic and navigation. We test our agent's abilities in text games -- challenging benchmarks for evaluating the multi-step reasoning abilities of game agents in grounded, language-based environments. Our experimental study indicates that injecting the actions from these symbolic modules into the action space of a behavior cloned transformer agent increases performance on four text game benchmarks that test a…
Transformer based models are the modern work horses for neural machine translation (NMT), reaching state of the art across several benchmarks. Despite their impressive accuracy, we observe a systemic and rudimentary class of errors made by transformer based models with regards to translating from a language that doesn't mark gender on nouns into others that do. We find that even when the surrounding context provides unambiguous evidence of the appropriate grammatical gender marking, no transformer based model we tested was able to accurately gender occupation nouns systematically. We release a…
Active learning has been shown to be an effective way to alleviate some of the effort required in utilising large collections of unlabelled data for machine learning tasks without needing to fully label them. The representation mechanism used to represent text documents when performing active learning, however, has a significant influence on how effective the process will be. While simple vector representations such as bag-of-words and embedding-based representations based on techniques such as word2vec have been shown to be an effective way to represent documents during active learning, the e…
Data and Web Science Group, University of Mannheim（曼海姆大学数据与网络科学组）；Wluper；Ubiquitous Knowledge Processing (UKP) Lab, TU Darmstadt（达姆施塔特工业大学普适知识处理（UKP）实验室）
Following the major success of neural language models (LMs) such as BERT or GPT-2 on a variety of language understanding tasks, recent work focused on injecting (structured) knowledge from external resources into these models. While on the one hand, joint pretraining (i.e., training from scratch, adding objectives based on external knowledge to the primary LM objective) may be prohibitively computationally expensive, post-hoc fine-tuning on external knowledge, on the other hand, may lead to the catastrophic forgetting of distributional knowledge. In this work, we investigate models for complem…
There have been a lot of interest in the scaling properties of Transformer models. However, not much has been done on the front of investigating the effect of scaling properties of different inductive biases and model architectures. Do model architectures scale differently? If so, how does inductive bias affect scaling behaviour? How does this influence upstream (pretraining) and downstream (transfer)? This paper conducts a systematic study of scaling behaviour of ten diverse model architectures such as Transformers, Switch Transformers, Universal Transformers, Dynamic convolutions, Performers…
As online news has become increasingly popular and fake news increasingly prevalent, the ability to audit the veracity of online news content has become more important than ever. Such a task represents a binary classification challenge, for which transformers have achieved state-of-the-art results. Using the publicly available ISOT and Combined Corpus datasets, this study explores transformers' abilities to identify fake news, with particular attention given to investigating generalisation to unseen datasets with varying styles, topics and class distributions. Moreover, we explore the idea tha…
Czech is a very specific language due to its large differences between the formal and the colloquial form of speech. While the formal (written) form is used mainly in official documents, literature, and public speeches, the colloquial (spoken) form is used widely among people in casual speeches. This gap introduces serious problems for ASR systems, especially when training or evaluating ASR models on datasets containing a lot of colloquial speech, such as the MALACH project. In this paper, we are addressing this problem in the light of a new paradigm in end-to-end ASR systems -- recently intro…
The huge size of the widely used BERT family models has led to recent efforts about model distillation. The main goal of distillation is to create a task-agnostic pre-trained model that can be fine-tuned on downstream tasks without fine-tuning its full-sized version. Despite the progress of distillation, to what degree and for what reason a task-agnostic model can be created from distillation has not been well studied. Also, the mechanisms behind transfer learning of those BERT models are not well investigated either. Therefore, this work focuses on analyzing the acceptable deduction when dist…
