This is an internal, incomplete preview of a proposed change to the ACL Anthology.
For efficiency reasons, we generate only three BibTeX files per volume, and the preview may be incomplete in other ways, or contain mistakes.
Do not treat this content as an official publication.
Fine-tuning large language models (LLMs) with domain-specific instruction dataset has emerged as an effective method to enhance their domain-specific understanding. Yet, there is limited work that examines the core characteristics acquired during this process. In this study, we benchmark the fundamental characteristics learned by contact-center (CC) domain specific instruction fine-tuned LLMs with out-of-the-box (OOB) LLMs via probing tasks encompassing conversational, channel, and automatic speech recognition (ASR) properties. We explore different LLM architectures (Flan-T5 and Llama) and sizes (3B, 7B, 11B, 13B). Our findings reveal remarkable effectiveness of CC-LLMs on the in-domain downstream tasks, with improvement in response acceptability by over 48% compared to OOB-LLMs. However, we observe that the performance of probing classifiers are relatively similar and does not reflect the performance of in-domain downstream tasks. A similar observation is also noted on SentEval dataset that assess capabilities of models in terms of surface, syntactic, and semantic information through probing tasks. Our study challenges the premise that probing classifiers can reveal the fundamental characteristics learned by large language models and is reflective of the downstream task performance, via a case-study of LLMs tuned for contact center domain.
This research investigates the impact of preference annotation acquisition methods on the performance of LLM alignment algorithms, including Direct Preference Optimization (DPO), Identity Preference Optimization (IPO), and Conservative DPO (cDPO), compared to Supervised Fine-Tuning (SFT) in NLP tasks. We analyze the influence of LLM and human-based preferences on algorithm performance, considering data volume and quality. Additionally, we assess DPO’s vulnerability to overfitting and IPO’s resilience against it, addressing four main research questions. Using the GAIR dataset and Zephyr-7b as the SFT model, we reveal unexpected negative outcomes. Specifically, DPO trained on LLM preferences outperforms human preferences, contrary to expectations. Moreover, there’s no correlation between preference data volume or quality and algorithm performance. Contrary to expectations, DPO shows no overfitting in both human and LLM preference datasets. Surprisingly, cDPO doesn’t fare better than DPO under flip noise. Our findings highlight the complexities of preference annotation methods and underscore the importance of scrutinizing negative results in NLP algorithm research.
In the dynamic realm of call center communications, the potential of abstractive summarization to transform information condensation is evident. However, evaluating the performance of abstractive summarization systems within contact center domain poses a significant challenge. Traditional evaluation metrics prove inadequate in capturing the multifaceted nature of call center conversations, characterized by diverse topics, emotional nuances, and dynamic contexts. This paper uses domain-specific perturbed summaries to scrutinize the robustness of summarization metrics in the call center domain. Through extensive experiments on call center data, we illustrate how perturbed summaries uncover limitations in existing metrics. We additionally utilize perturbation as data augmentation strategy to train domain-specific metrics. Our findings underscore the potential of perturbed summaries to complement current evaluation techniques, advancing reliable and adaptable summarization solutions in the call center domain.
Contact centers handle both chat and voice calls for the same domain. As part of their workflow, it is a standard practice to summarize the conversations once they conclude. A significant distinction between chat and voice communication lies in the presence of disfluencies in voice calls, such as repetitions, restarts, and replacements. These disfluencies are generally considered noise for downstream natural language understanding (NLU) tasks. While a separate summarization model for voice calls can be trained in addition to chat specific model for the same domain, it requires manual annotations for both the channels and adds complexity arising due to maintaining two models. Therefore, it’s crucial to investigate if a model trained on fluent data can handle disfluent data effectively. While previous research explored impact of disfluency on question-answering and intent detection, its influence on summarization is inadequately studied. Our experiments reveal up to 6.99-point degradation in Rouge-L score, along with reduced fluency, consistency, and relevance when a fluent-trained model handles disfluent data. Replacement disfluencies have the highest negative impact. To mitigate this, we examine Fused-Fine Tuning by training the model with a combination of fluent and disfluent data, resulting in improved performance on both public and real-life datasets. Our work highlights the significance of incorporating disfluency in training summarization models and its advantages in an industrial setting.
Transformer based language models have been widely adopted by industrial and research organisations in developing machine learning applications in the presence of limited annotated data. While these models show remarkable results, their functioning in few-shot settings is still poorly understood. Hence, we perform an investigative study to understand the characteristics of such models fine-tuned in few-shot setups. Specifically, we compare the intermediate layer representations obtained from a few-shot model and a pre-trained language model. We observe that pre-trained and few-shot models show similar representations over initial layers, whereas the later layers show a stark deviation. Based on these observations, we propose to freeze the initial Transformer layers to fine-tune the model in a constrained text classification setup with K annotated data points per class, where K ranges from 8 to 64. In our experiments across six benchmark sentence classification tasks, we discover that freezing initial 50% Transformer layers not only reduces training time but also surprisingly improves Macro F1 (upto 8%) when compared to fully trainable layers in few-shot setup. We also observe that this idea of layer freezing can very well be generalized to state-of-the-art few-shot text classification techniques, like DNNC and LM-BFF, leading to significant reduction in training time while maintaining comparable performance.
Weak Supervised Learning (WSL) is a popular technique to develop machine learning models in absence of labeled training data. WSL involves training over noisy labels which are traditionally obtained from hand-engineered semantic rules and task-specific pre-trained models. Such rules offer limited coverage and generalization over tasks. On the other hand, pre-trained models are available only for limited tasks. Thus, obtaining weak labels is a bottleneck in weak supervised learning. In this work, we propose to utilize the prompting paradigm to generate weak labels for the underlying tasks. We show that task-agnostic prompts are generalizable and can be used to obtain noisy labels for different Spoken Language Understanding (SLU) tasks such as sentiment classification, disfluency detection and emotion classification. These prompts can additionally be updated with human-in-the-loop to add task-specific contexts, thus providing flexibility to design task-specific prompts. Our proposed WSL pipeline outperforms other competitive low-resource benchmarks on zero and few-shot learning by more than 4% on Macro-F1 and a conventional rule-based WSL baseline by more than 5% across all the benchmark datasets. We demonstrate that prompt-based methods save nearly 75% of time in a weak-supervised framework and generate more reliable labels for the above SLU tasks and thus can be used as a universal strategy to obtain weak labels.
Stock market investors debate and heavily discuss stock ideas, investing strategies, news and market movements on social media platforms. The discussions are significantly longer in length and require extensive domain expertise for understanding. In this paper, we curate such discussions and construct a first-of-its-kind of abstractive summarization dataset. Our curated dataset consists of 7888 Reddit posts and manually constructed summaries for 400 posts. We robustly evaluate the summaries and conduct experiments on SOTA summarization tools to showcase their limitations. We plan to make the dataset publicly available. The sample dataset is available here: https://dhyeyjani.github.io/RSMC
Language Models (LMs) have been ubiquitously leveraged in various tasks including spoken language understanding (SLU). Spoken language requires careful understanding of speaker interactions, dialog states and speech induced multimodal behaviors to generate a meaningful representation of the conversation. In this work, we propose to dissect SLU into three representative properties: conversational (disfluency, pause, overtalk), channel (speaker-type, turn-tasks) and ASR (insertion, deletion, substitution). We probe BERT based language models (BERT, RoBERTa) trained on spoken transcripts to investigate its ability to understand multifarious properties in absence of any speech cues. Empirical results indicate that LM is surprisingly good at capturing conversational properties such as pause prediction and overtalk detection from lexical tokens. On the downsides, the LM scores low on turn-tasks and ASR errors predictions. Additionally, pre-training the LM on spoken transcripts restrain its linguistic understanding. Finally, we establish the efficacy and transferability of the mentioned properties on two benchmark datasets: Switchboard Dialog Act and Disfluency datasets.
Sentiment Analysis of code-mixed text has diversified applications in opinion mining ranging from tagging user reviews to identifying social or political sentiments of a sub-population. In this paper, we present an ensemble architecture of convolutional neural net (CNN) and self-attention based LSTM for sentiment analysis of code-mixed tweets. While the CNN component helps in the classification of positive and negative tweets, the self-attention based LSTM, helps in the classification of neutral tweets, because of its ability to identify correct sentiment among multiple sentiment bearing units. We achieved F1 scores of 0.707 (ranked 5th) and 0.725 (ranked 13th) on Hindi-English (Hinglish) and Spanish-English (Spanglish) datasets, respectively. The submissions for Hinglish and Spanglish tasks were made under the usernames ayushk and harsh_6 respectively.
In this paper, we propose a novel hybrid deep learning archtecture which is highly efficient for sentiment analysis in resource-poor languages. We learn sentiment embedded vectors from the Convolutional Neural Network (CNN). These are augmented to a set of optimized features selected through a multi-objective optimization (MOO) framework. The sentiment augmented optimized vector obtained at the end is used for the training of SVM for sentiment classification. We evaluate our proposed approach for coarse-grained (i.e. sentence level) as well as fine-grained (i.e. aspect level) sentiment analysis on four Hindi datasets covering varying domains. In order to show that our proposed method is generic in nature we also evaluate it on two benchmark English datasets. Evaluation shows that the results of the proposed method are consistent across all the datasets and often outperforms the state-of-art systems. To the best of our knowledge, this is the very first attempt where such a deep learning model is used for less-resourced languages such as Hindi.
This paper proposes OCR++, an open-source framework designed for a variety of information extraction tasks from scholarly articles including metadata (title, author names, affiliation and e-mail), structure (section headings and body text, table and figure headings, URLs and footnotes) and bibliography (citation instances and references). We analyze a diverse set of scientific articles written in English to understand generic writing patterns and formulate rules to develop this hybrid framework. Extensive evaluations show that the proposed framework outperforms the existing state-of-the-art tools by a large margin in structural information extraction along with improved performance in metadata and bibliography extraction tasks, both in terms of accuracy (around 50% improvement) and processing time (around 52% improvement). A user experience study conducted with the help of 30 researchers reveals that the researchers found this system to be very helpful. As an additional objective, we discuss two novel use cases including automatically extracting links to public datasets from the proceedings, which would further accelerate the advancement in digital libraries. The result of the framework can be exported as a whole into structured TEI-encoded documents. Our framework is accessible online at http://www.cnergres.iitkgp.ac.in/OCR++/home/.