"Flex Tape Can't Fix That": Bias and Misinformation in Edited Language Models

Model editing has emerged as a cost-effective strategy to update knowledge stored in language models. However, model editing can have unintended consequences after edits are applied: information unrelated to the edits can also be changed, and other general behaviors of the model can be wrongly altered. In this work, we investigate how model editing methods unexpectedly amplify model biases post-edit. We introduce a novel benchmark dataset, Seesaw-CF, for measuring bias-related harms of model editing and conduct the first in-depth investigation of how different weight-editing methods impact model bias. Specifically, we focus on biases with respect to demographic attributes such as race, geographic origin, and gender, as well as qualitative flaws in long-form texts generated by edited language models. We find that edited models exhibit, to various degrees, more biased behavior as they become less confident in attributes for Asian, African, and South American subjects. Furthermore, edited models amplify sexism and xenophobia in text generations while remaining seemingly coherent and logical. Finally, editing facts about place of birth, country of citizenship, or gender have particularly negative effects on the model's knowledge about unrelated features like field of work.

Investigating Cultural Alignment of Large Language Models

The intricate relationship between language and culture has long been a subject of exploration within the realm of linguistic anthropology. Large Language Models (LLMs), promoted as repositories of collective human knowledge, raise a pivotal question: do these models genuinely encapsulate the diverse knowledge adopted by different cultures? Our study reveals that these models demonstrate greater cultural alignment along two dimensions -- firstly, when prompted with the dominant language of a specific culture, and secondly, when pretrained with a refined mixture of languages employed by that culture. We quantify cultural alignment by simulating sociological surveys, comparing model responses to those of actual survey participants as references. Specifically, we replicate a survey conducted in various regions of Egypt and the United States through prompting LLMs with different pretraining data mixtures in both Arabic and English with the personas of the real respondents and the survey questions. Further analysis reveals that misalignment becomes more pronounced for underrepresented personas and for culturally sensitive topics, such as those probing social values. Finally, we introduce Anthropological Prompting, a novel method leveraging anthropological reasoning to enhance cultural alignment. Our study emphasizes the necessity for a more balanced multilingual pretraining dataset to better represent the diversity of human experience and the plurality of different cultures with many implications on the topic of cross-lingual transfer.

Partial Diacritization: A Context-Contrastive Inference Approach

Diacritization plays a pivotal role in improving readability and disambiguating the meaning of Arabic texts. Efforts have so far focused on marking every eligible character (Full Diacritization). Comparatively overlooked, Partial Diacritzation (PD) is the selection of a subset of characters to be marked to aid comprehension where needed. Research has indicated that excessive diacritic marks can hinder skilled readers--reducing reading speed and accuracy. We conduct a behavioral experiment and show that partially marked text is often easier to read than fully marked text, and sometimes easier than plain text. In this light, we introduce Context-Contrastive Partial Diacritization (CCPD)--a novel approach to PD which integrates seamlessly with existing Arabic diacritization systems. CCPD processes each word twice, once with context and once without, and diacritizes only the characters with disparities between the two inferences. Further, we introduce novel indicators for measuring partial diacritization quality (SR, PDER, HDER, ERE), essential for establishing this as a machine learning task. Lastly, we introduce TD2, a Transformer-variant of an established model which offers a markedly different performance profile on our proposed indicators compared to all other known systems.

Instruction-tuning Aligns LLMs to the Human Brain

Instruction-tuning is a widely adopted method of finetuning that enables large language models (LLMs) to generate output that more closely resembles human responses to natural language queries, in many cases leading to human-level performance on diverse testbeds. However, it remains unclear whether instruction-tuning truly makes LLMs more similar to how humans process language. We investigate the effect of instruction-tuning on LLM-human similarity in two ways: (1) brain alignment, the similarity of LLM internal representations to neural activity in the human language system, and (2) behavioral alignment, the similarity of LLM and human behavior on a reading task. We assess 25 vanilla and instruction-tuned LLMs across three datasets involving humans reading naturalistic stories and sentences. We discover that instruction-tuning generally enhances brain alignment by an average of 6%, but does not have a similar effect on behavioral alignment. To identify the factors underlying LLM-brain alignment, we compute correlations between the brain alignment of LLMs and various model properties, such as model size, various problem-solving abilities, and performance on tasks requiring world knowledge spanning various domains. Notably, we find a strong positive correlation between brain alignment and model size (r = 0.95), as well as performance on tasks requiring world knowledge (r = 0.81). Our results demonstrate that instruction-tuning LLMs improves both world knowledge representations and brain alignment, suggesting that mechanisms that encode world knowledge in LLMs also improve representational alignment to the human brain.

Rosetta Stone at KSAA-RD Shared Task: A Hop From Language Modeling To Word--Definition Alignment

A Reverse Dictionary is a tool enabling users to discover a word based on its provided definition, meaning, or description. Such a technique proves valuable in various scenarios, aiding language learners who possess a description of a word without its identity, and benefiting writers seeking precise terminology. These scenarios often encapsulate what is referred to as the "Tip-of-the-Tongue" (TOT) phenomena. In this work, we present our winning solution for the Arabic Reverse Dictionary shared task. This task focuses on deriving a vector representation of an Arabic word from its accompanying description. The shared task encompasses two distinct subtasks: the first involves an Arabic definition as input, while the second employs an English definition. For the first subtask, our approach relies on an ensemble of finetuned Arabic BERT-based models, predicting the word embedding for a given definition. The final representation is obtained through averaging the output embeddings from each model within the ensemble. In contrast, the most effective solution for the second subtask involves translating the English test definitions into Arabic and applying them to the finetuned models originally trained for the first subtask. This straightforward method achieves the highest score across both subtasks.

Taqyim: Evaluating Arabic NLP Tasks Using ChatGPT Models

Large language models (LLMs) have demonstrated impressive performance on various downstream tasks without requiring fine-tuning, including ChatGPT, a chat-based model built on top of LLMs such as GPT-3.5 and GPT-4. Despite having a lower training proportion compared to English, these models also exhibit remarkable capabilities in other languages. In this study, we assess the performance of GPT-3.5 and GPT-4 models on seven distinct Arabic NLP tasks: sentiment analysis, translation, transliteration, paraphrasing, part of speech tagging, summarization, and diacritization. Our findings reveal that GPT-4 outperforms GPT-3.5 on five out of the seven tasks. Furthermore, we conduct an extensive analysis of the sentiment analysis task, providing insights into how LLMs achieve exceptional results on a challenging dialectal dataset. Additionally, we introduce a new Python interface https://github.com/ARBML/Taqyim that facilitates the evaluation of these tasks effortlessly.

OPT-R: Exploring the Role of Explanations in Finetuning and Prompting for Reasoning Skills of Large Language Models

In this paper, we conduct a thorough investigation into the reasoning capabilities of Large Language Models (LLMs), focusing specifically on the Open Pretrained Transformers (OPT) models as a representative of such models. Our study entails finetuning three different sizes of OPT on a carefully curated reasoning corpus, resulting in two sets of finetuned models: OPT-R, finetuned without explanations, and OPT-RE, finetuned with explanations. We then evaluate all models on 57 out-of-domain tasks drawn from the SUPER-NATURALINSTRUCTIONS benchmark, covering 26 distinct reasoning skills, utilizing three prompting techniques. Through a comprehensive grid of 27 configurations and 6,156 test evaluations, we investigate the dimensions of finetuning, prompting, and scale to understand the role of explanations on different reasoning skills. Our findings reveal that having explanations in the fewshot exemplar has no significant impact on the model's performance when the model is finetuned, while positively affecting the non-finetuned counterpart. Moreover, we observe a slight yet consistent increase in classification accuracy as we incorporate explanations during prompting and finetuning, respectively. Finally, we offer insights on which skills benefit the most from incorporating explanations during finetuning and prompting, such as Numerical (+20.4%) and Analogical (+13.9%) reasoning, as well as skills that exhibit negligible or negative effects.

Depth-Wise Attention (DWAtt): A Layer Fusion Method for Data-Efficient Classification

Language Models pretrained on large textual data have been shown to encode different types of knowledge simultaneously. Traditionally, only the features from the last layer are used when adapting to new tasks or data. We put forward that, when using or finetuning deep pretrained models, intermediate layer features that may be relevant to the downstream task are buried too deep to be used efficiently in terms of needed samples or steps. To test this, we propose a new layer fusion method: Depth-Wise Attention (DWAtt), to help re-surface signals from non-final layers. We compare DWAtt to a basic concatenation-based layer fusion method (Concat), and compare both to a deeper model baseline -- all kept within a similar parameter budget. Our findings show that DWAtt and Concat are more step- and sample-efficient than the baseline, especially in the few-shot setting. DWAtt outperforms Concat on larger data sizes. On CoNLL-03 NER, layer fusion shows 3.68-9.73% F1 gain at different few-shot sizes. The layer fusion models presented significantly outperform the baseline in various training scenarios with different data sizes, architectures, and training constraints.

ToKen: Task Decomposition and Knowledge Infusion for Few-Shot Hate Speech Detection

Hate speech detection is complex; it relies on commonsense reasoning, knowledge of stereotypes, and an understanding of social nuance that differs from one culture to the next. It is also difficult to collect a large-scale hate speech annotated dataset. In this work, we frame this problem as a few-shot learning task, and show significant gains with decomposing the task into its "constituent" parts. In addition, we see that infusing knowledge from reasoning datasets (e.g. Atomic2020) improves the performance even further. Moreover, we observe that the trained models generalize to out-of-distribution datasets, showing the superiority of task decomposition and knowledge infusion compared to previously used methods. Concretely, our method outperforms the baseline by 17.83% absolute gain in the 16-shot case.

Meta AI at Arabic Hate Speech 2022: MultiTask Learning with Self-Correction for Hate Speech Classification

In this paper, we tackle the Arabic Fine-Grained Hate Speech Detection shared task and demonstrate significant improvements over reported baselines for its three subtasks. The tasks are to predict if a tweet contains (1) Offensive language; and whether it is considered (2) Hate Speech or not and if so, then predict the (3) Fine-Grained Hate Speech label from one of six categories. Our final solution is an ensemble of models that employs multitask learning and a self-consistency correction method yielding 82.7% on the hate speech subtask -- reflecting a 3.4% relative improvement compared to previous work.