Computation and Language (cs.CL)

Abstract: The mixing of two or more languages is called Code-Mixing (CM). CM is a social norm in multilingual societies. Neural Language Models (NLMs) like transformers have been very effective on many NLP tasks. However, NLM for CM is an under-explored area. Though transformers are capable and powerful, they cannot always encode positional/sequential information since they are non-recurrent. Therefore, to enrich word information and incorporate positional information, positional encoding is defined. We hypothesize that Switching Points (SPs), i.e., junctions in the text where the language switches (L1 -> L2 or L2-> L1), pose a challenge for CM Language Models (LMs), and hence give special emphasis to switching points in the modeling process. We experiment with several positional encoding mechanisms and show that rotatory positional encodings along with switching point information yield the best results. We introduce CONFLATOR: a neural language modeling approach for code-mixed languages. CONFLATOR tries to learn to emphasize switching points using smarter positional encoding, both at unigram and bigram levels. CONFLATOR outperforms the state-of-the-art on two tasks based on code-mixed Hindi and English (Hinglish): (i) sentiment analysis and (ii) machine translation.

Abstract: Many meetings require creating a meeting summary to keep everyone up to date. Creating minutes of sufficient quality is however very cognitively demanding. Although we currently possess capable models for both audio speech recognition (ASR) and summarization, their fully automatic use is still problematic. ASR models frequently commit errors when transcribing named entities while the summarization models tend to hallucinate and misinterpret the transcript. We propose a novel tool -- Minuteman -- to enable efficient semi-automatic meeting minuting. The tool provides a live transcript and a live meeting summary to the users, who can edit them in a collaborative manner, enabling correction of ASR errors and imperfect summary points in real time. The resulting application eases the cognitive load of the notetakers and allows them to easily catch up if they missed a part of the meeting due to absence or a lack of focus. We conduct several tests of the application in varied settings, exploring the worthiness of the concept and the possible user strategies.

Abstract: Transfer learning has led to large gains in performance for nearly all NLP tasks while making downstream models easier and faster to train. This has also been extended to low-resourced languages, with some success. We investigate the properties of cross-lingual transfer learning between ten low-resourced languages, from the perspective of a named entity recognition task. We specifically investigate how much adaptive fine-tuning and the choice of transfer language affect zero-shot transfer performance. We find that models that perform well on a single language often do so at the expense of generalising to others, while models with the best generalisation to other languages suffer in individual language performance. Furthermore, the amount of data overlap between the source and target datasets is a better predictor of transfer performance than either the geographical or genetic distance between the languages.

Abstract: This paper summarizes the results of experimenting with Universal Dependencies (UD) adaptation of an Unsupervised, Compositional and Recursive (UCR) rule-based approach for Sentiment Analysis (SA) submitted to the Shared Task at Rest-Mex 2023 (Team Olga/LyS-SALSA) (within the IberLEF 2023 conference). By using basic syntactic rules such as rules of modification and negation applied on words from sentiment dictionaries, our approach exploits some advantages of an unsupervised method for SA: (1) interpretability and explainability of SA, (2) robustness across datasets, languages and domains and (3) usability by non-experts in NLP. We compare our approach with other unsupervised approaches of SA that in contrast to our UCR rule-based approach use simple heuristic rules to deal with negation and modification. Our results show a considerable improvement over these approaches. We discuss future improvements of our results by using modality features as another shifting rule of polarity and word disambiguation techniques to identify the right sentiment words.

Abstract: Transformer-based Language Models are widely used in Natural Language Processing related tasks. Thanks to their pre-training, they have been successfully adapted to Information Extraction in business documents. However, most pre-training tasks proposed in the literature for business documents are too generic and not sufficient to learn more complex structures. In this paper, we use LayoutLM, a language model pre-trained on a collection of business documents, and introduce two new pre-training tasks that further improve its capacity to extract relevant information. The first is aimed at better understanding the complex layout of documents, and the second focuses on numeric values and their order of magnitude. These tasks force the model to learn better-contextualized representations of the scanned documents. We further introduce a new post-processing algorithm to decode BIESO tags in Information Extraction that performs better with complex entities. Our method significantly improves extraction performance on both public (from 93.88 to 95.50 F1 score) and private (from 84.35 to 84.84 F1 score) datasets composed of expense receipts, invoices, and purchase orders.

Abstract: The Mixture of Experts (MoE) is a widely known neural architecture where an ensemble of specialized sub-models optimizes overall performance with a constant computational cost. However, conventional MoEs pose challenges at scale due to the need to store all experts in memory. In this paper, we push MoE to the limit. We propose extremely parameter-efficient MoE by uniquely combining MoE architecture with lightweight experts.Our MoE architecture outperforms standard parameter-efficient fine-tuning (PEFT) methods and is on par with full fine-tuning by only updating the lightweight experts -- less than 1% of an 11B parameters model. Furthermore, our method generalizes to unseen tasks as it does not depend on any prior task knowledge. Our research underscores the versatility of the mixture of experts architecture, showcasing its ability to deliver robust performance even when subjected to rigorous parameter constraints. Our code used in all the experiments is publicly available here: https://github.com/for-ai/parameter-efficient-moe.

Abstract: High-quality instruction-tuning data is critical to improving LLM capabilities. Existing data collection methods are limited by unrealistic manual labeling costs or by the hallucination of relying solely on LLM generation. To address the problems, this paper presents a scalable method to automatically collect high-quality instructional adaptation data by training language models to automatically design tasks based on human-written texts. Intuitively, human-written text helps to help the model attenuate illusions during the generation of tasks. Unlike instruction back-translation-based methods that directly take the given text as a response, we require the model to generate the \textit{instruction}, \textit{input}, and \textit{output} simultaneously to filter the noise. The results of the automated and manual evaluation experiments demonstrate the quality of our dataset.

Abstract: The development of highly fluent large language models (LLMs) has prompted increased interest in assessing their reasoning and problem-solving capabilities. We investigate whether several LLMs can solve a classic type of deductive reasoning problem from the cognitive science literature. The tested LLMs have limited abilities to solve these problems in their conventional form. We performed follow up experiments to investigate if changes to the presentation format and content improve model performance. We do find performance differences between conditions; however, they do not improve overall performance. Moreover, we find that performance interacts with presentation format and content in unexpected ways that differ from human performance. Overall, our results suggest that LLMs have unique reasoning biases that are only partially predicted from human reasoning performance.

Abstract: Readability metrics and standards such as Flesch Kincaid Grade Level (FKGL) and the Common European Framework of Reference for Languages (CEFR) exist to guide teachers and educators to properly assess the complexity of educational materials before administering them for classroom use. In this study, we select a diverse set of open and closed-source instruction-tuned language models and investigate their performances in writing story completions and simplifying narratives$-$tasks that teachers perform$-$using standard-guided prompts controlling text readability. Our extensive findings provide empirical proof of how globally recognized models like ChatGPT may be considered less effective and may require more refined prompts for these generative tasks compared to other open-sourced models such as BLOOMZ and FlanT5$-$which have shown promising results.

Abstract: We continue the investigation into the power of smaller Transformer-based language models as initiated by \textbf{TinyStories} -- a 10 million parameter model that can produce coherent English -- and the follow-up work on \textbf{phi-1}, a 1.3 billion parameter model with Python coding performance close to the state-of-the-art. The latter work proposed to use existing Large Language Models (LLMs) to generate ``textbook quality" data as a way to enhance the learning process compared to traditional web data. We follow the ``Textbooks Are All You Need" approach, focusing this time on common sense reasoning in natural language, and create a new 1.3 billion parameter model named \textbf{phi-1.5}, with performance on natural language tasks comparable to models 5x larger, and surpassing most non-frontier LLMs on more complex reasoning tasks such as grade-school mathematics and basic coding. More generally, \textbf{phi-1.5} exhibits many of the traits of much larger LLMs, both good -- such as the ability to ``think step by step" or perform some rudimentary in-context learning -- and bad, including hallucinations and the potential for toxic and biased generations -- encouragingly though, we are seeing improvement on that front thanks to the absence of web data. We open-source \textbf{phi-1.5} to promote further research on these urgent topics.

Abstract: Self-supervised learning (SSL) is at the origin of unprecedented improvements in many different domains including computer vision and natural language processing. Speech processing drastically benefitted from SSL as most of the current domain-related tasks are now being approached with pre-trained models. This work introduces LeBenchmark 2.0 an open-source framework for assessing and building SSL-equipped French speech technologies. It includes documented, large-scale and heterogeneous corpora with up to 14,000 hours of heterogeneous speech, ten pre-trained SSL wav2vec 2.0 models containing from 26 million to one billion learnable parameters shared with the community, and an evaluation protocol made of six downstream tasks to complement existing benchmarks. LeBenchmark 2.0 also presents unique perspectives on pre-trained SSL models for speech with the investigation of frozen versus fine-tuned downstream models, task-agnostic versus task-specific pre-trained models as well as a discussion on the carbon footprint of large-scale model training.

Abstract: Demographics, Social determinants of health, and family history documented in the unstructured text within the electronic health records are increasingly being studied to understand how this information can be utilized with the structured data to improve healthcare outcomes. After the GPT models were released, many studies have applied GPT models to extract this information from the narrative clinical notes. Different from the existing work, our research focuses on investigating the zero-shot learning on extracting this information together by providing minimum information to the GPT model. We utilize de-identified real-world clinical notes annotated for demographics, various social determinants, and family history information. Given that the GPT model might provide text different from the text in the original data, we explore two sets of evaluation metrics, including the traditional NER evaluation metrics and semantic similarity evaluation metrics, to completely understand the performance. Our results show that the GPT-3.5 method achieved an average of 0.975 F1 on demographics extraction, 0.615 F1 on social determinants extraction, and 0.722 F1 on family history extraction. We believe these results can be further improved through model fine-tuning or few-shots learning. Through the case studies, we also identified the limitations of the GPT models, which need to be addressed in future research.

Abstract: Social media platforms play an essential role in crisis communication, but analyzing crisis-related social media texts is challenging due to their informal nature. Transformer-based pre-trained models like BERT and RoBERTa have shown success in various NLP tasks, but they are not tailored for crisis-related texts. Furthermore, general-purpose sentence encoders are used to generate sentence embeddings, regardless of the textual complexities in crisis-related texts. Advances in applications like text classification, semantic search, and clustering contribute to effective processing of crisis-related texts, which is essential for emergency responders to gain a comprehensive view of a crisis event, whether historical or real-time. To address these gaps in crisis informatics literature, this study introduces CrisisTransformers, an ensemble of pre-trained language models and sentence encoders trained on an extensive corpus of over 15 billion word tokens from tweets associated with more than 30 crisis events, including disease outbreaks, natural disasters, conflicts, and other critical incidents. We evaluate existing models and CrisisTransformers on 18 crisis-specific public datasets. Our pre-trained models outperform strong baselines across all datasets in classification tasks, and our best-performing sentence encoder improves the state-of-the-art by 17.43% in sentence encoding tasks. Additionally, we investigate the impact of model initialization on convergence and evaluate the significance of domain-specific models in generating semantically meaningful sentence embeddings. All models are publicly released (https://huggingface.co/crisistransformers), with the anticipation that they will serve as a robust baseline for tasks involving the analysis of crisis-related social media texts.

Abstract: Personality types are important in various fields as they hold relevant information about the characteristics of a human being in an explainable format. They are often good predictors of a person's behaviors in a particular environment and have applications ranging from candidate selection to marketing and mental health. Recently automatic detection of personality traits from texts has gained significant attention in computational linguistics. Most personality detection and analysis methods have focused on small datasets making their experimental observations often limited. To bridge this gap, we focus on collecting and releasing the largest automatically curated dataset for the research community which has 152 million tweets and 56 thousand data points for the Myers-Briggs personality type (MBTI) prediction task. We perform a series of extensive qualitative and quantitative studies on our dataset to analyze the data patterns in a better way and infer conclusions. We show how our intriguing analysis results often follow natural intuition. We also perform a series of ablation studies to show how the baselines perform for our dataset.

Abstract: In recent years, natural language processing has gained significant popularity in various sectors, including the legal domain. This paper presents NeCo Team's solutions to the Vietnamese text processing tasks provided in the Automated Legal Question Answering Competition 2023 (ALQAC 2023), focusing on legal domain knowledge acquisition for low-resource languages through data enrichment. Our methods for the legal document retrieval task employ a combination of similarity ranking and deep learning models, while for the second task, which requires extracting an answer from a relevant legal article in response to a question, we propose a range of adaptive techniques to handle different question types. Our approaches achieve outstanding results on both tasks of the competition, demonstrating the potential benefits and effectiveness of question answering systems in the legal field, particularly for low-resource languages.

Abstract: The evolution of Generative Pre-trained Transformer (GPT) models has led to significant advancements in various natural language processing applications, particularly in legal textual entailment. We present an analysis of GPT-3.5 (ChatGPT) and GPT-4 performances on COLIEE Task 4 dataset, a prominent benchmark in this domain. The study encompasses data from Heisei 18 (2006) to Reiwa 3 (2021), exploring the models' abilities to discern entailment relationships within Japanese statute law across different periods. Our preliminary experimental results unveil intriguing insights into the models' strengths and weaknesses in handling legal textual entailment tasks, as well as the patterns observed in model performance. In the context of proprietary models with undisclosed architectures and weights, black-box analysis becomes crucial for evaluating their capabilities. We discuss the influence of training data distribution and the implications on the models' generalizability. This analysis serves as a foundation for future research, aiming to optimize GPT-based models and enable their successful adoption in legal information extraction and entailment applications.

Abstract: Since their release, Transformers have revolutionized many fields from Natural Language Understanding to Computer Vision. Document Understanding (DU) was not left behind with first Transformer based models for DU dating from late 2019. However, the computational complexity of the self-attention operation limits their capabilities to small sequences. In this paper we explore multiple strategies to apply Transformer based models to long multi-page documents. We introduce 2 new multi-modal (text + layout) long-range models for DU. They are based on efficient implementations of Transformers for long sequences. Long-range models can process whole documents at once effectively and are less impaired by the document's length. We compare them to LayoutLM, a classical Transformer adapted for DU and pre-trained on millions of documents. We further propose 2D relative attention bias to guide self-attention towards relevant tokens without harming model efficiency. We observe improvements on multi-page business documents on Information Retrieval for a small performance cost on smaller sequences. Relative 2D attention revealed to be effective on dense text for both normal and long-range models.

Abstract: Large Language Models (LLMs) have proven their exceptional capabilities in performing language-related tasks. However, their deployment poses significant challenges due to their considerable memory and storage requirements. In response to this issue, weight-only quantization, particularly 3 and 4-bit weight-only quantization, has emerged as one of the most viable solutions. As the number of bits decreases, the quantization grid broadens, thus emphasizing the importance of up and down rounding. While previous studies have demonstrated that fine-tuning up and down rounding with the addition of perturbations can enhance accuracy in some scenarios, our study is driven by the precise and limited boundary of these perturbations, where only the threshold for altering the rounding value is of significance. Consequently, we propose a concise and highly effective approach for optimizing the weight rounding task. Our method, named SignRound, involves lightweight block-wise tuning using signed gradient descent, enabling us to achieve outstanding results within 400 steps. SignRound outperforms the established baseline of rounding-to-nearest (RTN) and competes impressively against recent methods, without introducing additional inference overhead. The source code will be publicly available at https://github.com/intel/neural-compressor soon.

Abstract: Text-to-image synthesis for the Chinese language poses unique challenges due to its large vocabulary size, and intricate character relationships. While existing diffusion models have shown promise in generating images from textual descriptions, they often neglect domain-specific contexts and lack robustness in handling the Chinese language. This paper introduces PAI-Diffusion, a comprehensive framework that addresses these limitations. PAI-Diffusion incorporates both general and domain-specific Chinese diffusion models, enabling the generation of contextually relevant images. It explores the potential of using LoRA and ControlNet for fine-grained image style transfer and image editing, empowering users with enhanced control over image generation. Moreover, PAI-Diffusion seamlessly integrates with Alibaba Cloud's Machine Learning Platform for AI, providing accessible and scalable solutions. All the Chinese diffusion model checkpoints, LoRAs, and ControlNets, including domain-specific ones, are publicly available. A user-friendly Chinese WebUI and the diffusers-api elastic inference toolkit, also open-sourced, further facilitate the easy deployment of PAI-Diffusion models in various environments, making it a valuable resource for Chinese text-to-image synthesis.

Abstract: Large language models (LLMs) can respond to human language queries and have shown powerful potential applications in network operations (NetOps). Thanks to the large amount of commonsense knowledge inherent, LLMs achieve much better inference accuracy than traditional models and emerge with strong abilities in generalization, reasoning, and code generation. These abilities may have a crucial boost to automated and intelligent NetOps. However, it remains under-explored how well LLMs perform in various NetOps tasks. In this work, we make a systematic assessment of the capabilities, strengths, and limitations of selected LLMs in the field of NetOps. The evaluation is conducted on a collection of 5,732 questions about NetOps, encompassing 26 publicly available general-domain LLMs, including ChatGPT, LLaMA, Falcon, etc. We also finetune some of these LLMs with our collected NetOps corpus and evaluate the resulting models. The evaluation method follows the widely adopted benchmarks for general-domain LLMs, combined with Chain-of-Thought Prompts and Retrieval-Augmented Generation. The results show that only GPT-4 achieves high accuracy equivalent to passing the NetOps certification exam for humans, while all the other LLMs have much lower accuracy. However, some open models like LLaMA 2 still demonstrate significant potential. Furthermore, we evaluate the impact of factors such as model parameters, prompt engineering, instruction fine-tuning etc. This work shall be treated as the initial effort to systematic evaluation of LLMs in NetOps, and a more rigorous study is required for production use. The evaluation code and dataset will be released to benefit future research.

Abstract: Answering multi-hop reasoning questions requires retrieving and synthesizing information from diverse sources. Large Language Models (LLMs) struggle to perform such reasoning consistently. Here we propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on LLM attention heads. First, we analyze the per-layer activations of GPT-2 models in response to single and multi-hop prompts. We then propose a mechanism that allows users to inject pertinent prompt-specific information, which we refer to as "memories," at critical LLM locations during inference. By thus enabling the LLM to incorporate additional relevant information during inference, we enhance the quality of multi-hop prompt completions. We show empirically that a simple, efficient, and targeted memory injection into a key attention layer can often increase the probability of the desired next token in multi-hop tasks, by up to 424%.

Abstract: Multimodal stock trading volume movement prediction with stock-related news is one of the fundamental problems in the financial area. Existing multimodal works that train models from scratch face the problem of lacking universal knowledge when modeling financial news. In addition, the models ability may be limited by the lack of domain-related knowledge due to insufficient data in the datasets. To handle this issue, we propose the Prompt-based MUltimodal Stock volumE prediction model (ProMUSE) to process text and time series modalities. We use pre-trained language models for better comprehension of financial news and adopt prompt learning methods to leverage their capability in universal knowledge to model textual information. Besides, simply fusing two modalities can cause harm to the unimodal representations. Thus, we propose a novel cross-modality contrastive alignment while reserving the unimodal heads beside the fusion head to mitigate this problem. Extensive experiments demonstrate that our proposed ProMUSE outperforms existing baselines. Comprehensive analyses further validate the effectiveness of our architecture compared to potential variants and learning mechanisms.

Abstract: Large language models (LLMs) have demonstrated significant capability to generalize across a large number of NLP tasks. For industry applications, it is imperative to assess the performance of the LLM on unlabeled production data from time to time to validate for a real-world setting. Human labeling to assess model error requires considerable expense and time delay. Here we demonstrate that ensemble disagreement scores work well as a proxy for human labeling for language models in zero-shot, few-shot, and fine-tuned settings, per our evaluation on keyphrase extraction (KPE) task. We measure fidelity of the results by comparing to true error measured from human labeled ground truth. We contrast with the alternative of using another LLM as a source of machine labels, or silver labels. Results across various languages and domains show disagreement scores provide a better estimation of model performance with mean average error (MAE) as low as 0.4% and on average 13.8% better than using silver labels.

Abstract: We introduce MAmmoTH, a series of open-source large language models (LLMs) specifically tailored for general math problem-solving. The MAmmoTH models are trained on MathInstruct, our meticulously curated instruction tuning dataset. MathInstruct is compiled from 13 math datasets with intermediate rationales, six of which have rationales newly curated by us. It presents a unique hybrid of chain-of-thought (CoT) and program-of-thought (PoT) rationales, and also ensures extensive coverage of diverse fields in math. The hybrid of CoT and PoT not only unleashes the potential of tool use but also allows different thought processes for different math problems. As a result, the MAmmoTH series substantially outperform existing open-source models on nine mathematical reasoning datasets across all scales with an average accuracy gain between 13% and 29%. Remarkably, our MAmmoTH-7B model reaches 35% on MATH (a competition-level dataset), which exceeds the best open-source 7B model (WizardMath) by 25%, and the MAmmoTH-34B model achieves 46% accuracy on MATH, even surpassing GPT-4's CoT result. Our work underscores the importance of diverse problem coverage and the use of hybrid rationales in developing superior math generalist models.