We introduce a frustratingly simple, highly efficient, and surprisingly effective decoding method, termed Frustratingly Simple Decoding (FSD), for neural text generation. The idea behind FSD is straightforward: We construct an anti-language model (anti-LM) based on previously generated text, which is employed to penalize the future generation of repetitive content. The anti-LM can be implemented as simple as an n-gram language model or a vectorized variant. In this way, FSD incurs no additional model parameters and negligible computational overhead (FSD can be as fast as greedy search). Despite its simplicity, FSD is surprisingly effective and generalizes across different datasets, models, and languages. Extensive experiments show that FSD outperforms established strong baselines in terms of generation quality, decoding speed, and universality.

While Large Language Models (LLMs) have demonstrated exceptional multitasking abilities, fine-tuning these models on downstream, domain-specific datasets is often necessary to yield superior performance on test sets compared to their counterparts without fine-tuning. However, the comprehensive effects of fine-tuning on the LLMs’ generalization ability are not fully understood.This paper delves into the differences between original, unmodified LLMs and their fine-tuned variants. Our primary investigation centers on whether fine-tuning affects the generalization ability intrinsic to LLMs. To elaborate on this, we conduct extensive experiments across five distinct language tasks on various datasets.Our main findings reveal that models fine-tuned on generation and classification tasks exhibit dissimilar behaviors in generalizing to different domains and tasks.Intriguingly, we observe that integrating the in-context learning strategy during fine-tuning on generation tasks can enhance the model’s generalization ability.Through this systematic investigation, we aim to contribute valuable insights into the evolving landscape of fine-tuning practices for LLMs.

In the realm of emerging multitasking abilities of Large language models (LLMs), methodologies like prompt tuning enable low-cost adaptation to downstream tasks without retraining the model. However, automatic input pre-processing when LLMs are unavailable is currently under-studied. This paper proposes ReLLM (Rephrasing for LLMs), a method that automatically paraphrases input content for better output generations. ReLLM replaces low-frequency lexical items with their high-frequency counterparts. This substitution is particularly beneficial for low-resource language tasks that lack sufficient training data and resources. ReLLM is user-friendly and requires no additional LLM training. Experimental results in cross-lingual summarization, and natural language inference demonstrate the effectiveness of ReLLM.

In this paper, we study the generalization ability of large language models (LLMs) with respect to compositional instructions, which are instructions that can be decomposed into several sub-instructions. We argue that the ability to generalize from simple instructions to more intricate compositional instructions represents a key aspect of the out-of-distribution generalization for LLMs. Since there are no specialized datasets for studying this phenomenon, we first construct a dataset with the help of ChatGPT, guided by the self-instruct technique. Then, we fine-tune and evaluate LLMs on these datasets. Interestingly, our experimental results indicate that training LLMs on higher-order compositional instructions enhances their performance on lower-order ones, but the reverse does not hold true.

Commonsense generation aims to generate a plausible sentence containing all given unordered concept words. Previous methods focusing on this task usually directly concatenate these words as the input of a pre-trained language model (PLM). However, in PLMs’ pre-training process, the inputs are often corrupted sentences with correct word order. This input distribution discrepancy between pre-training and fine-tuning makes the model difficult to fully utilize the knowledge of PLMs. In this paper, we propose a two-stage framework to alleviate this issue. Firstly, in pre-training stage, we design a new format of input to endow PLMs the ability to deal with masked sentences with incorrect word order. Secondly, during fine-tuning, we insert the special token [MASK] between two consecutive concept words to make the input distribution more similar to the input distribution in pre-training. We conduct extensive experiments and provide thorough analysis to demonstrate the effectiveness of our proposed method.

Exemplar-Guided Paraphrase Generation (EGPG) aims to generate a target sentence which conforms to the style of the given exemplar while encapsulating the content information of the source sentence. In this paper, we propose a new method with the goal of learning a better representation of the style and the content. This method is mainly motivated by the recent success of contrastive learning which has demonstrated its power in unsupervised feature extraction tasks. The idea is to design two contrastive losses with respect to the content and the style by considering two problem characteristics during training. One characteristic is that the target sentence shares the same content with the source sentence, and the second characteristic is that the target sentence shares the same style with the exemplar. These two contrastive losses are incorporated into the general encoder-decoder paradigm. Experiments on two datasets, namely QQP-Pos and ParaNMT, demonstrate the effectiveness of our proposed constrastive losses.

Emphasis Selection is a newly proposed task which focuses on choosing words for emphasis in short sentences. Traditional methods only consider the sequence information of a sentence while ignoring the rich sentence structure and word relationship information. In this paper, we propose a new framework that considers sentence structure via a sentence structure graph and word relationship via a word similarity graph. The sentence structure graph is derived from the parse tree of a sentence. The word similarity graph allows nodes to share information with their neighbors since we argue that in emphasis selection, similar words are more likely to be emphasized together. Graph neural networks are employed to learn the representation of each node of these two graphs. Experimental results demonstrate that our framework can achieve superior performance.