Canada

Is the output softmax layer, which is adopted by most language models (LMs), always the best way to compute the next word probability? Given so many attention layers in a modern transformer-based LM, are the pointer networks redundant nowadays? In this study, we discover that the answers to both questions are no. This is because the softmax bottleneck sometimes prevents the LMs from predicting the desired distribution and the pointer networks can be used to break the bottleneck efficiently. Based on the finding, we propose several softmax alternatives by simplifying the pointer networks and accelerating the word-by-word rerankers. In GPT-2, our proposals are significantly better and more efficient than mixture of softmax, a state-of-the-art softmax alternative. In summarization experiments, without significantly decreasing its training/testing speed, our best method based on T5-Small improves factCC score by 2 points in CNN/DM and XSUM dataset, and improves MAUVE scores by 30% in BookSum paragraph-level dataset.

ACL 2023

Revisiting the Architectures like Pointer Networks to Efficiently Improve the Next Word Distribution, Summarization Factuality, and Beyond

next word distribution

reranker

mixture of softmax

softmax bottleneck

repetition

copy mechanism

hallucination

bart

factuality

gpt-2

summarization

pointer network

### Welcome to ACL 2023, the 61st Annual Meeting of the Association for Computational Linguistics! 
 The conference will be held in Toronto, Canada, July 9-14, 2023. 
Following the succession of the recent conferences in our field, ACL 2023 will adopt a hybrid format.
While the impact of Covid has considerably diminished in terms of traveling, obtaining visas to Canada
entails a very long process. Moreover, the global economic conditions pose challenges for many individuals to travel to conferences. Recognizing these circumstances, we know many participants may not be
able to attend the conference in person. Therefore, we are committed to providing a great virtual platform
so everyone has the opportunity to interact with other participants and enjoy the conference. Based on the
current registered participants, approxiately 30% have chosen to attend the conference virtually. Whether
you join us in person or virtually, we sincerely hope everyone has a remarkable conference experience. 
This General Chair’s message is where I express my gratitude to the many individuals who have made
enormous contributions to the conference over the past year.

Read [**ACL 2023 General Chair's message**](https://docs.google.com/document/d/1WobYM7norbG4dI48s75HfJoD89qgX5a_F-6U8AteLSA/edit?usp=sharing/) in full.

##### **[Conference Handbook](https://2023.aclweb.org/downloads/acl2023-handbook.pdf)**

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing.

workshop paper

Byte-Pair Encoding (BPE) is a popular algorithm used for tokenizing data in NLP, despite being devised initially as a compression method. BPE appears to be a greedy algorithm at face value, but the underlying optimization problem that BPE seeks to solve has not yet been laid down. We formalize BPE as a combinatorial optimization problem. Via submodular functions, we prove that the iterative greedy version is a 1/σ (1-e^(-σ))$-approximation of an optimal merge sequence, where σ is the total backward curvature with respect to the optimal merge sequence. Empirically the lower bound of the approximation is ≈0.37.

We provide a faster implementation of BPE which improves the runtime complexity from O(N M) to O(N log M), where N is the sequence length and M is the merge count. Finally, we optimize the brute-force algorithm for optimal BPE using memoization.

Byte-Pair Encoding is Approximately Optimal

Pretrained language models have demonstrated extraordinary capabilities in language generation. However, real-world tasks often require controlling the distribution of generated text in order to mitigate bias, promote fairness, and achieve personalization. Existing techniques for controlling the distribution of generated text only work with quantified distributions, which require pre-defined categories, proportions of the distribution, or an existing corpus following the desired distributions. However, many important distributions, such as personal preferences, are unquantified. In this work, we tackle the problem of generating text following arbitrary distributions (quantified and unquantified) by proposing NANO, a few-shot human-in-the-loop training algorithm that continuously learns from human feedback. NANO achieves state-of-the-art results on single topic/attribute as well as quantified distribution control compared to previous works. We also show that NANO is able to learn unquantified distributions, achieves personalization, and captures differences between different individuals' personal preferences with high sample efficiency.

Nano: Nested Human-in-the-Loop Reward Learning for Few-shot Language Model Control

The unprecedented rate at which machine learning (ML) models are growing in
size necessitates novel approaches to enable efficient and scalable solutions. We
contribute to this line of work by studying a novel version of the Budgeted Correlation Clustering problem (BCC) where along with a limited number of queries
to an expensive oracle for node similarities (e.g. a large ML model), we have
unlimited access to a cheaper but less accurate second oracle. Our formulation is
inspired by many practical scenarios where coarse approximations of the expen-
sive similarity metric can be efficiently obtained via weaker models. We develop
a theoretically motivated algorithm that leverages the cheap oracle to judiciously
query the strong oracle while maintaining high clustering quality. We empirically
demonstrate gains in query minimization and clustering metrics on a variety of
datasets with diverse strong and cheap oracles. Most notably, we demonstrate
a practical application in text clustering based on expensive cross-attention lan-
guage models by showing that cheaper (but weaker) embedding-based models can
be leveraged to substantially reduce the number of inference calls to the former.

KwikBucks: Correlation Clustering with Cheap-Weak and Expensive-Strong Signals
Sandeep Silwal, Sara Ahmadian, Andrew Nystrom, Andrew McCallum, Deepak Ramachandran and Mehran Kazemi (mehrankazemi@google.com)
Status: Accept

We introduce the range of oBERTa language models, an easy-to-use set
of language models which allows Natural Language Processing (NLP) practitioners to obtain between 3.8 and 24.3 times faster models without expertise in model compression. Specifically, oBERTa extends existing work on pruning, knowledge distillation, and quantization and leverages frozen embeddings, improves
distillation, and model initialization to deliver higher accuracy on a broad range of transfer tasks.

oBERTa: Improving Sparse Transfer Learning via improved initialization, distillation, and pruning regimes
Daniel Campos, Alexandre Marques, Mark Kurtz and Cheng Xiang Zhai (dcampos3@illinois.edu)
Status: Accept

We consider the problem of improving the inference latency of language model-based dense retrieval systems by introducing structural compression and model size asymmetry between the context and query encoders. First, we investigate the impact of pre and post-training compression on the MSMARCO, Natural Questions, TriviaQA, SQUAD, and SCIFACT, finding that asymmetry in the dual-encoders in dense retrieval can lead to improved inference efficiency. Knowing this, we introduce Kullback–Leibler Alignment of Embeddings (KALE), an efficient and accurate method for increasing the inference efficiency of dense retrieval methods by pruning and aligning the query encoder after training. Specifically, KALE extends traditional Knowledge Distillation after bi-encoder training, allowing for effective query encoder compression without full retraining or index generation. Using KALE and asymmetric training, we can generate models which exceed the performance of DistilBERT despite having 3x faster inference.

Quick Dense Retrievers Consume KALE: Post Training Kullback–Leibler Alignment of Embeddings for Asymmetrical dual encoders
Daniel Campos, Alessandro Magnani and Chengxiang Zhai (dcampos3@illinois.edu)
Status: Accept

We propose a parameter sharing method for Transformers (Vaswani et al., 2017). The proposed approach relaxes a widely used technique, which shares parameters for one layer with all layers such as Universal Transformers (Dehghani et al., 2019), to increase the efficiency in the computational time. We propose three strategies: Sequence, Cycle, and Cycle (rev) to assign parameters to each layer. Experimental results show that the proposed strategies are efficient in the parameter size and computational time. Moreover, we indicate that the proposed strategies are also effective in the configuration where we use many training data such as the recent WMT competition.

Lessons on Parameter Sharing across Layers in Transformers

Sequence-to-sequence language models can be used to produce abstractive summaries which are coherent, relevant, and concise. Still, model sizes can make deployment in latency-sensitive or web-scale implementations difficult. This paper studies the relationship between model size, structured pruning, inference efficiency, and summarization accuracy on widely used summarization datasets. We show that model accuracy is tied to the encoder size while inference efficiency is connected to the decoder. Using asymmetric pruning can lead to nearly 3x improvement in inference latency with ~1 point loss in Rouge-2. Moreover, we find both the average degradation and the role of asymmetry to be consistent across model sizes and variations in datasets.

To Asymmetry and Beyond: Structured Pruning of Sequence to Sequence Models for Improved Inference Efficiency
Daniel Campos and ChengXiang Zhai (dcampos3@illinois.edu)
Status: Accept

We examine the effects of model size and pre- finetuning in an active learning setting where classifiers are trained from scratch on 14 binary and 3 multi-class text classification tasks. We make an important observation that, in realistic active learning settings, where the human anno- tator and the active learning system operate in asynchronous mode, a compact pre-finetuned 1-layer transformer model with 4.2 million pa- rameters is 30% more label efficient when com- pared to the larger 24-layer 84 million param- eter transformer model. Further, in line with previous studies, we note that pre-finetuning transformer models on related tasks improves label efficiency of downstream tasks by 12%- 50%. The compact pre-finetuned model does not require GPUs, making it a viable solution for large-scale real-time inference with cheaper CPU options.

Small is the New Big: Pre-finetuned compact models are better for Asynchronous Active Learning
Dantong Liu, Kaushik Pavani and Sunny Dasgupta (lidanton@amazon.com)
Status: Accept

Fine-tuning large pre-trained models for downstream tasks can be really expensive. In the past, researchers have proposed various alternatives like adapter and prompt-based methods for tuning these large language models using minimal parameters. However, applying prompt-tuning for smaller language models has not been effective so far and not much work is done in pushing forward soft prompting for
these smaller models. To improve the training efficiency of the language models and reduce the size of tuned parameters, we propose a novel Adapter-based Efficient Prompt Tuning approach (ADEPT). In this paper, we show that tuning the parameters of soft prompts with adapter modules while keeping the rest of the model frozen can be a promising method to optimize smaller language models for downstream tasks. Our method achieves up to 98% performance of full fine-tuning while using only 0.02% of total model parameters.

ADEPT: Adapter-based Efficient Prompt Tuning Approach for Language Models
Aditya Shah, Surendrabikram Thapa, Aneesh Jain and Lifu Huang (surendrabikram@vt.edu)
Status: Accept

Finetuning large language models inflates the costs of NLU applications and remains the bottleneck of development cycles. Recent works in computer vision use data pruning to reduce training time. Pruned data selection with static methods is based on a score calculated for each training example prior to finetuning, which involves important computational overhead. Moreover, the score may not necessarily be representative of sample importance throughout the entire training duration. We propose to address these issues with a refined version of dynamic data pruning, a curriculum which periodically scores and discards unimportant examples during finetuning. Our method leverages an EL2N metric that we extend to the joint intent and slot classification task, and an initial finetuning phase on the full train set. Our results on the GLUE benchmark and four joint NLU datasets show a better time-accuracy trade-off compared to static methods. Our method preserves full accuracy while training on 50% of the data points and reduces computational times by up to 41%. If we tolerate instead a minor drop of accuracy of 1%, we can prune 80% of the training examples for a reduction in finetuning time reaching 66%.

NLU on Data Diets: Dynamic Data Subset Selection for NLP Classification Tasks
Jean-Michel Attendu and Jean-Philippe Corbeil (jeanphilippecorbeil@gmail.com)
Status: Accept

Downloads

Next from ACL 2023

Byte-Pair Encoding is Approximately Optimal

Similar lecture

Revisiting the Architectures like Pointer Networks to Efficiently Improve the Next Word Distribution, Summarization Factuality, and Beyond