Computation and Language (cs.CL)

Abstract: Despite the many use cases for large language models (LLMs) in the design of chatbots in various industries and the research showing the importance of personalizing chatbots to cater to different personality traits, little work has been done to evaluate whether the behaviors of personalized LLMs can reflect certain personality traits accurately and consistently. We consider studying the behavior of LLM-based simulated agents which refer to as LLM personas and present a case study with GPT-3.5 (text-davinci-003) to investigate whether LLMs can generate content with consistent, personalized traits when assigned Big Five personality types and gender roles. We created 320 LLM personas (5 females and 5 males for each of the 32 Big Five personality types) and prompted them to complete the classic 44-item Big Five Inventory (BFI) and then write an 800-word story about their childhood. Results showed that LLM personas' self-reported BFI scores are consistent with their assigned personality types, with large effect sizes found on all five traits. Moreover, significant correlations were found between assigned personality types and some Linguistic Inquiry and Word Count (LIWC) psycholinguistic features of their writings. For instance, extroversion is associated with pro-social and active words, and neuroticism is associated with words related to negative emotions and mental health. Besides, we only found significant differences in using technological and cultural words in writing between LLM-generated female and male personas. This work provides a first step for further research on personalized LLMs and their applications in Human-AI conversation.

Abstract: The recent advent of self-supervised pre-training techniques has led to a surge in the use of multimodal learning in form document understanding. However, existing approaches that extend the mask language modeling to other modalities require careful multi-task tuning, complex reconstruction target designs, or additional pre-training data. In FormNetV2, we introduce a centralized multimodal graph contrastive learning strategy to unify self-supervised pre-training for all modalities in one loss. The graph contrastive objective maximizes the agreement of multimodal representations, providing a natural interplay for all modalities without special customization. In addition, we extract image features within the bounding box that joins a pair of tokens connected by a graph edge, capturing more targeted visual cues without loading a sophisticated and separately pre-trained image embedder. FormNetV2 establishes new state-of-the-art performance on FUNSD, CORD, SROIE and Payment benchmarks with a more compact model size.

Abstract: Although large language models demonstrate remarkable question-answering performances, revealing the intermediate reasoning steps that the models faithfully follow remains challenging. In this paper, we propose FAME (FAithful question answering with MontE-carlo planning) to answer questions based on faithful reasoning steps. The reasoning steps are organized as a structured entailment tree, which shows how premises are used to produce intermediate conclusions that can prove the correctness of the answer. We formulate the task as a discrete decision-making problem and solve it through the interaction of a reasoning environment and a controller. The environment is modular and contains several basic task-oriented modules, while the controller proposes actions to assemble the modules. Since the search space could be large, we introduce a Monte-Carlo planning algorithm to do a look-ahead search and select actions that will eventually lead to high-quality steps. FAME achieves state-of-the-art performance on the standard benchmark. It can produce valid and faithful reasoning steps compared with large language models with a much smaller model size.

Abstract: Analysis and extraction of useful information from legal judgments using computational linguistics was one of the earliest problems posed in the domain of information retrieval. Presently, several commercial vendors exist who automate such tasks. However, a crucial bottleneck arises in the form of exorbitant pricing and lack of resources available in analysis of judgements mete out by Hong Kong's Legal System. This paper attempts to bridge this gap by providing several statistical, machine learning, deep learning and zero-shot learning based methods to effectively analyze legal judgments from Hong Kong's Court System. The methods proposed consists of: (1) Citation Network Graph Generation, (2) PageRank Algorithm, (3) Keyword Analysis and Summarization, (4) Sentiment Polarity, and (5) Paragrah Classification, in order to be able to extract key insights from individual as well a group of judgments together. This would make the overall analysis of judgments in Hong Kong less tedious and more automated in order to extract insights quickly using fast inferencing. We also provide an analysis of our results by benchmarking our results using Large Language Models making robust use of the HuggingFace ecosystem.

Abstract: To better support information retrieval tasks such as web search and open-domain question answering, growing effort is made to develop retrieval-oriented language models, e.g., RetroMAE and many others. Most of the existing works focus on improving the semantic representation capability for the contextualized embedding of the [CLS] token. However, recent study shows that the ordinary tokens besides [CLS] may provide extra information, which help to produce a better representation effect. As such, it's necessary to extend the current methods where all contextualized embeddings can be jointly pre-trained for the retrieval tasks. In this work, we propose a novel pre-training method called Duplex Masked Auto-Encoder, a.k.a. DupMAE. It is designed to improve the quality of semantic representation where all contextualized embeddings of the pre-trained model can be leveraged. It takes advantage of two complementary auto-encoding tasks: one reconstructs the input sentence on top of the [CLS] embedding; the other one predicts the bag-of-words feature of the input sentence based on the ordinary tokens' embeddings. The two tasks are jointly conducted to train a unified encoder, where the whole contextualized embeddings are aggregated in a compact way to produce the final semantic representation. DupMAE is simple but empirically competitive: it substantially improves the pre-trained model's representation capability and transferability, where superior retrieval performances can be achieved on popular benchmarks, like MS MARCO and BEIR.

Abstract: Keyphrase prediction aims to generate phrases (keyphrases) that highly summarizes a given document. Recently, researchers have conducted in-depth studies on this task from various perspectives. In this paper, we comprehensively summarize representative studies from the perspectives of dominant models, datasets and evaluation metrics. Our work analyzes up to 167 previous works, achieving greater coverage of this task than previous surveys. Particularly, we focus highly on deep learning-based keyphrase prediction, which attracts increasing attention of this task in recent years. Afterwards, we conduct several groups of experiments to carefully compare representative models. To the best of our knowledge, our work is the first attempt to compare these models using the identical commonly-used datasets and evaluation metric, facilitating in-depth analyses of their disadvantages and advantages. Finally, we discuss the possible research directions of this task in the future.

Abstract: Large-scale open-domain dialogue data crawled from public social media has greatly improved the performance of dialogue models. However, long-turn dialogues are still highly scarce. Specifically, most dialogue sessions in existing corpora have less than three turns. To alleviate this issue, we propose the Retrieve, Reorganize and Rescale framework (Re$^3$Dial), which can automatically construct a billion-scale long-turn dialogue corpus from existing short-turn dialogue data. Re$^3$Dial first trains an Unsupervised Dense Session Retriever (UDSR) to capture semantic and discourse relationships within multi-turn dialogues for retrieving relevant and coherent sessions. It then reorganizes the short-turn dialogues into long-turn sessions via recursively retrieving and selecting the consecutive sessions with our proposed diversity sampling strategy. Extensive evaluations on multiple multi-turn dialogue benchmarks demonstrate that Re$^3$Dial consistently and significantly improves the dialogue model's ability to utilize long-term context for modeling multi-turn dialogues across different pre-training settings. Finally, we build a toolkit for efficiently rescaling dialogue corpus with Re$^3$Dial, which enables us to construct a corpus containing 1B Chinese dialogue sessions with 11.3 turns on average (5X longer than the original EVA corpus). We will release our UDSR model, toolkit, and data for public use.

Abstract: In recent years, sentiment analysis has gained significant importance in natural language processing. However, most existing models and datasets for sentiment analysis are developed for high-resource languages, such as English and Chinese, leaving low-resource languages, particularly African languages, largely unexplored. The AfriSenti-SemEval 2023 Shared Task 12 aims to fill this gap by evaluating sentiment analysis models on low-resource African languages. In this paper, we present our solution to the shared task, where we employed different multilingual XLM-R models with classification head trained on various data, including those retrained in African dialects and fine-tuned on target languages. Our team achieved the third-best results in Subtask B, Track 16: Multilingual, demonstrating the effectiveness of our approach. While our model showed relatively good results on multilingual data, it performed poorly in some languages. Our findings highlight the importance of developing more comprehensive datasets and models for low-resource African languages to advance sentiment analysis research. We also provided the solution on the github repository.

Abstract: The affective reasoning task is a set of emerging affect-based tasks in conversation, including Emotion Recognition in Conversation (ERC),Emotion-Cause Pair Extraction (ECPE), and Emotion-Cause Span Recognition (ECSR). Existing methods make various assumptions on the apparent relationship while neglecting the essential causal model due to the nonuniqueness of skeletons and unobservability of implicit causes. This paper settled down the above two problems and further proposed Conversational Affective Causal Discovery (CACD). It is a novel causal discovery method showing how to discover causal relationships in a conversation via designing a common skeleton and generating a substitute for implicit causes. CACD contains two steps: (i) building a common centering one graph node causal skeleton for all utterances in variable-length conversations; (ii) Causal Auto-Encoder (CAE) correcting the skeleton to yield causal representation through generated implicit causes and known explicit causes. Comprehensive experiments demonstrate that our novel method significantly outperforms the SOTA baselines in six affect-related datasets on the three tasks.

Abstract: Artificial neural networks (ANNs) are increasingly used as research models, but questions remain about their generalizability and representational invariance. Biological neural networks under social constraints evolved to enable communicable representations, demonstrating generalization capabilities. This study proposes a communication protocol between cooperative agents to analyze the formation of individual and shared abstractions and their impact on task performance. This communication protocol aims to mimic language features by encoding high-dimensional information through low-dimensional representation. Using grid-world mazes and reinforcement learning, teacher ANNs pass a compressed message to a student ANN for better task completion. Through this, the student achieves a higher goal-finding rate and generalizes the goal location across task worlds. Further optimizing message content to maximize student reward improves information encoding, suggesting that an accurate representation in the space of messages requires bi-directional input. This highlights the role of language as a common representation between agents and its implications on generalization capabilities.

Abstract: Language models generate text based on successively sampling the next word. A decoding procedure based on nucleus (top-$p$) sampling chooses from the smallest possible set of words whose cumulative probability exceeds the probability $p$. In this work, we assess whether a top-$p$ set is indeed aligned with its probabilistic meaning in various linguistic contexts. We employ conformal prediction, a calibration procedure that focuses on the construction of minimal prediction sets according to a desired confidence level, to calibrate the parameter $p$ as a function of the entropy of the next word distribution. We find that OPT models are overconfident, and that calibration shows a moderate inverse scaling with model size.

Abstract: As pointed out by several scholars, current research on hate speech (HS) recognition is characterized by unsystematic data creation strategies and diverging annotation schemata. Subsequently, supervised-learning models tend to generalize poorly to datasets they were not trained on, and the performance of the models trained on datasets labeled using different HS taxonomies cannot be compared. To ease this problem, we propose applying extremely weak supervision that only relies on the class name rather than on class samples from the annotated data. We demonstrate the effectiveness of a state-of-the-art weakly-supervised text classification model in various in-dataset and cross-dataset settings. Furthermore, we conduct an in-depth quantitative and qualitative analysis of the source of poor generalizability of HS classification models.

Abstract: Multilingual Machine Translation promises to improve translation quality between non-English languages. This is advantageous for several reasons, namely lower latency (no need to translate twice), and reduced error cascades (e.g., avoiding losing gender and formality information when translating through English). On the downside, adding more languages reduces model capacity per language, which is usually countered by increasing the overall model size, making training harder and inference slower. In this work, we introduce Language-Specific Transformer Layers (LSLs), which allow us to increase model capacity, while keeping the amount of computation and the number of parameters used in the forward pass constant. The key idea is to have some layers of the encoder be source or target language-specific, while keeping the remaining layers shared. We study the best way to place these layers using a neural architecture search inspired approach, and achieve an improvement of 1.3 chrF (1.5 spBLEU) points over not using LSLs on a separate decoder architecture, and 1.9 chrF (2.2 spBLEU) on a shared decoder one.

Abstract: Word-level saliency explanations ("heat maps over words") are often used to communicate feature-attribution in text-based models. Recent studies found that superficial factors such as word length can distort human interpretation of the communicated saliency scores. We conduct a user study to investigate how the marking of a word's neighboring words affect the explainee's perception of the word's importance in the context of a saliency explanation. We find that neighboring words have significant effects on the word's importance rating. Concretely, we identify that the influence changes based on neighboring direction (left vs. right) and a-priori linguistic and computational measures of phrases and collocations (vs. unrelated neighboring words). Our results question whether text-based saliency explanations should be continued to be communicated at word level, and inform future research on alternative saliency explanation methods.

Abstract: Reinforcement learning has been applied to train the dialog systems in many works. Previous approaches divide the dialog system into multiple modules including DST (dialog state tracking) and DP (dialog policy), and train these modules simultaneously. However, different modules influence each other during training. The errors from DST might misguide the dialog policy, and the system action brings extra difficulties for the DST module. To alleviate this problem, we propose Asynchronous Updating Reinforcement Learning framework (AURL) that updates the DST module and the DP module asynchronously under a cooperative setting. Furthermore, curriculum learning is implemented to address the problem of unbalanced data distribution during reinforcement learning sampling, and multiple user models are introduced to increase the dialog diversity. Results on the public SSD-PHONE dataset show that our method achieves a compelling result with a 31.37% improvement on the dialog success rate. The code is publicly available via https://github.com/shunjiu/AURL.

Abstract: Dialogue Topic Segmentation (DTS) plays an essential role in a variety of dialogue modeling tasks. Previous DTS methods either focus on semantic similarity or dialogue coherence to assess topic similarity for unsupervised dialogue segmentation. However, the topic similarity cannot be fully identified via semantic similarity or dialogue coherence. In addition, the unlabeled dialogue data, which contains useful clues of utterance relationships, remains underexploited. In this paper, we propose a novel unsupervised DTS framework, which learns topic-aware utterance representations from unlabeled dialogue data through neighboring utterance matching and pseudo-segmentation. Extensive experiments on two benchmark datasets (i.e., DialSeg711 and Doc2Dial) demonstrate that our method significantly outperforms the strong baseline methods. For reproducibility, we provide our code and data at:https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/dial-start.

Abstract: Proactive dialogue systems, related to a wide range of real-world conversational applications, equip the conversational agent with the capability of leading the conversation direction towards achieving pre-defined targets or fulfilling certain goals from the system side. It is empowered by advanced techniques to progress to more complicated tasks that require strategical and motivational interactions. In this survey, we provide a comprehensive overview of the prominent problems and advanced designs for conversational agent's proactivity in different types of dialogues. Furthermore, we discuss challenges that meet the real-world application needs but require a greater research focus in the future. We hope that this first survey of proactive dialogue systems can provide the community with a quick access and an overall picture to this practical problem, and stimulate more progresses on conversational AI to the next level.

Abstract: Existing neural machine translation (NMT) studies mainly focus on developing dataset-specific models based on data from different tasks (e.g., document translation and chat translation). Although the dataset-specific models have achieved impressive performance, it is cumbersome as each dataset demands a model to be designed, trained, and stored. In this work, we aim to unify these translation tasks into a more general setting. Specifically, we propose a ``versatile'' model, i.e., the Unified Model Learning for NMT (UMLNMT) that works with data from different tasks, and can translate well in multiple settings simultaneously, and theoretically it can be as many as possible. Through unified learning, UMLNMT is able to jointly train across multiple tasks, implementing intelligent on-demand translation. On seven widely-used translation tasks, including sentence translation, document translation, and chat translation, our UMLNMT results in substantial improvements over dataset-specific models with significantly reduced model deployment costs. Furthermore, UMLNMT can achieve competitive or better performance than state-of-the-art dataset-specific methods. Human evaluation and in-depth analysis also demonstrate the superiority of our approach on generating diverse and high-quality translations. Additionally, we provide a new genre translation dataset about famous aphorisms with 186k Chinese->English sentence pairs.

Abstract: Recent advances in deep learning methods for natural language processing (NLP) have created new business opportunities and made NLP research critical for industry development. As one of the big players in the field of NLP, together with governments and universities, it is important to track the influence of industry on research. In this study, we seek to quantify and characterize industry presence in the NLP community over time. Using a corpus with comprehensive metadata of 78,187 NLP publications and 701 resumes of NLP publication authors, we explore the industry presence in the field since the early 90s. We find that industry presence among NLP authors has been steady before a steep increase over the past five years (180% growth from 2017 to 2022). A few companies account for most of the publications and provide funding to academic researchers through grants and internships. Our study shows that the presence and impact of the industry on natural language processing research are significant and fast-growing. This work calls for increased transparency of industry influence in the field.

Abstract: In this thesis, we develop methods to enhance the interpretability of recent representation learning techniques in natural language processing (NLP) while accounting for the unavailability of annotated data. We choose to leverage Variational Autoencoders (VAEs) due to their efficiency in relating observations to latent generative factors and their effectiveness in data-efficient learning and interpretable representation learning. As a first contribution, we identify and remove unnecessary components in the functioning scheme of semi-supervised VAEs making them faster, smaller and easier to design. Our second and main contribution is to use VAEs and Transformers to build two models with inductive bias to separate information in latent representations into understandable concepts without annotated data. The first model, Attention-Driven VAE (ADVAE), is able to separately represent and control information about syntactic roles in sentences. The second model, QKVAE, uses separate latent variables to form keys and values for its Transformer decoder and is able to separate syntactic and semantic information in its neural representations. In transfer experiments, QKVAE has competitive performance compared to supervised models and equivalent performance to a supervised model using 50K annotated samples. Additionally, QKVAE displays improved syntactic role disentanglement capabilities compared to ADVAE. Overall, we demonstrate that it is possible to enhance the interpretability of state-of-the-art deep learning architectures for language modeling with unannotated data in situations where text data is abundant but annotations are scarce.

Abstract: Controlling chatbot utterance generation with multiple attributes such as personalities, emotions and dialogue acts is a practically useful but under-studied problem. We propose a novel controllable generation framework called DASC that possesses strong controllability with weighted decoding paradigm, while improving generation quality with the grounding in an attribute semantics space. Generation with multiple attributes is then intuitively implemented with an interpolation of multiple attribute embeddings. Experiments show that DASC can achieve state-of-the-art control accuracy in 3-aspect controllable generation tasks while also producing interesting and reasonably sensible responses, even if in an out-of-distribution robustness test. Visualization of the meaningful representations learned in the attribute semantic space also supports its effectiveness.

Abstract: Domain shift is a big challenge in NLP, thus, many approaches resort to learning domain-invariant features to mitigate the inference phase domain shift. Such methods, however, fail to leverage the domain-specific nuances relevant to the task at hand. To avoid such drawbacks, domain counterfactual generation aims to transform a text from the source domain to a given target domain. However, due to the limited availability of data, such frequency-based methods often miss and lead to some valid and spurious domain-token associations. Hence, we employ a three-step domain obfuscation approach that involves frequency and attention norm-based masking, to mask domain-specific cues, and unmasking to regain the domain generic context. Our experiments empirically show that the counterfactual samples sourced from our masked text lead to improved domain transfer on 10 out of 12 domain sentiment classification settings, with an average of 2% accuracy improvement over the state-of-the-art for unsupervised domain adaptation (UDA). Further, our model outperforms the state-of-the-art by achieving 1.4% average accuracy improvement in the adversarial domain adaptation (ADA) setting. Moreover, our model also shows its domain adaptation efficacy on a large multi-domain intent classification dataset where it attains state-of-the-art results. We release the codes publicly at \url{https://github.com/declare-lab/remask}.

Abstract: Dataset bias, i.e., the over-reliance on dataset-specific literal heuristics, is getting increasing attention for its detrimental effect on the generalization ability of NLU models. Existing works focus on eliminating dataset bias by down-weighting problematic data in the training process, which induce the omission of valid feature information while mitigating bias. In this work, We analyze the causes of dataset bias from the perspective of causal inference and propose CausalAPM, a generalizable literal disentangling framework to ameliorate the bias problem from feature granularity. The proposed approach projects literal and semantic information into independent feature subspaces, and constrains the involvement of literal information in subsequent predictions. Extensive experiments on three NLP benchmarks (MNLI, FEVER, and QQP) demonstrate that our proposed framework significantly improves the OOD generalization performance while maintaining ID performance.

Abstract: Acceleration of large language model pre-training is a critical issue in present NLP research. In this paper, we focus on speeding up pre-training by progressively growing from a small Transformer structure to a large one. There are two main research problems related to progressive growth: growth schedule and growth operator. For growth schedule, existing work has explored multi-stage expansion of depth and feedforward layers. However, the impact of each dimension on the schedule's efficiency is still an open question. For growth operator, existing work relies on the initialization of new weights to inherit knowledge, and achieve only non-strict function preservation, limiting further optimization of training dynamics. To address these issues, we propose Masked Structural Growth (MSG), including growth schedules involving all possible dimensions and strictly function-preserving growth operators that is independent of the initialization of new weights. Experiments show that MSG is significantly faster than related work: we achieve a speed-up of 80% for Bert-base and 120% for Bert-large pre-training. Moreover, MSG is able to improve fine-tuning performances at the same time.

Abstract: Emergent chain-of-thought (CoT) reasoning capabilities promise to improve performance and explainability of large language models (LLMs). However, uncertainties remain about how prompting strategies formulated for previous model generations generalize to new model generations and different datasets. In this small-scale study we compare the performance of a range of zero-shot prompts for inducing CoT reasoning across six recently released LLMs (davinci-002, davinci-003, GPT-3.5-turbo, GPT-4, Flan-T5-xxl and Cohere command-xlarge) on a mixture of six question-answering datasets, including datasets from scientific and medical domains. We find that a CoT prompt that was previously discovered through automated prompt discovery shows robust performance across experimental conditions and produces best results when applied to the state-of-the-art model GPT-4.

Abstract: It is challenging to extract semantic meanings directly from audio signals in spoken language understanding (SLU), due to the lack of textual information. Popular end-to-end (E2E) SLU models utilize sequence-to-sequence automatic speech recognition (ASR) models to extract textual embeddings as input to infer semantics, which, however, require computationally expensive auto-regressive decoding. In this work, we leverage self-supervised acoustic encoders fine-tuned with Connectionist Temporal Classification (CTC) to extract textual embeddings and use joint CTC and SLU losses for utterance-level SLU tasks. Experiments show that our model achieves 4% absolute improvement over the the state-of-the-art (SOTA) dialogue act classification model on the DSTC2 dataset and 1.3% absolute improvement over the SOTA SLU model on the SLURP dataset.

Abstract: This paper describes the results of SemEval 2023 task 7 -- Multi-Evidence Natural Language Inference for Clinical Trial Data (NLI4CT) -- consisting of 2 tasks, a Natural Language Inference (NLI) task, and an evidence selection task on clinical trial data. The proposed challenges require multi-hop biomedical and numerical reasoning, which are of significant importance to the development of systems capable of large-scale interpretation and retrieval of medical evidence, to provide personalized evidence-based care. Task 1, the entailment task, received 643 submissions from 40 participants, and Task 2, the evidence selection task, received 364 submissions from 23 participants. The tasks are challenging, with the majority of submitted systems failing to significantly outperform the majority class baseline on the entailment task, and we observe significantly better performance on the evidence selection task than on the entailment task. Increasing the number of model parameters leads to a direct increase in performance, far more significant than the effect of biomedical pre-training. Future works could explore the limitations of large models for generalization and numerical inference, and investigate methods to augment clinical datasets to allow for more rigorous testing and to facilitate fine-tuning. We envisage that the dataset, models, and results of this task will be useful to the biomedical NLI and evidence retrieval communities. The dataset, competition leaderboard, and website are publicly available.

Abstract: Despite growing interest in applications based on natural customer support conversations, there exist remarkably few publicly available datasets that reflect the expected characteristics of conversations in these settings. Existing task-oriented dialogue datasets, which were collected to benchmark dialogue systems mainly in written human-to-bot settings, are not representative of real customer support conversations and do not provide realistic benchmarks for systems that are applied to natural data. To address this gap, we introduce NatCS, a multi-domain collection of spoken customer service conversations. We describe our process for collecting synthetic conversations between customers and agents based on natural language phenomena observed in real conversations. Compared to previous dialogue datasets, the conversations collected with our approach are more representative of real human-to-human conversations along multiple metrics. Finally, we demonstrate potential uses of NatCS, including dialogue act classification and intent induction from conversations as potential applications, showing that dialogue act annotations in NatCS provide more effective training data for modeling real conversations compared to existing synthetic written datasets. We publicly release NatCS to facilitate research in natural dialog systems

Abstract: Sentence-level representations are beneficial for various natural language processing tasks. It is commonly believed that vector representations can capture rich linguistic properties. Currently, large language models (LMs) achieve state-of-the-art performance on sentence embedding. However, some recent works suggest that vector representations from LMs can cause information leakage. In this work, we further investigate the information leakage issue and propose a generative embedding inversion attack (GEIA) that aims to reconstruct input sequences based only on their sentence embeddings. Given the black-box access to a language model, we treat sentence embeddings as initial tokens' representations and train or fine-tune a powerful decoder model to decode the whole sequences directly. We conduct extensive experiments to demonstrate that our generative inversion attack outperforms previous embedding inversion attacks in classification metrics and generates coherent and contextually similar sentences as the original inputs.

Abstract: This project focuses on enhancing open-source large language models through instruction-tuning and providing comprehensive evaluations of their performance. We explore how various training data factors, such as quantity, quality, and linguistic distribution, influence the performance of instruction-tuned models trained on publicly accessible high-quality instruction datasets for both English and Chinese languages. Our goal is to supplement evaluation with quantitative analyses, providing valuable insights for the continued advancement of open-source chat models. Our model, data, and code are publicly available for others to use and build upon.

Abstract: We introduce three simple randomized variants of byte pair encoding (BPE) and explore whether randomizing the selection of merge operations substantially affects a downstream machine translation task. We focus on translation into morphologically rich languages, hypothesizing that this task may show sensitivity to the method of choosing subwords. Analysis using a Bayesian linear model indicates that two of the variants perform nearly indistinguishably compared to standard BPE while the other degrades performance less than we anticipated. We conclude that although standard BPE is widely used, there exists an interesting universe of potential variations on it worth investigating. Our code is available at: https://github.com/bltlab/random-bpe.