Croatia

Novel intent discovery automates the process of grouping similar messages (questions) to identify previously unknown intents. However, current research focuses on publicly available datasets which have only the question field and significantly differ from real-life datasets. This paper proposes methods to improve the intent discovery pipeline deployed in a large e-commerce platform. We show the benefit of pre-training language models on in-domain data: both self-supervised and with weak supervision. We also devise the best method to utilize the conversational structure (i.e., question and answer) of real-life datasets during fine-tuning for clustering tasks, which we call Conv. All our methods combined to fully utilize real-life datasets give up to 33pp performance boost over state-of-the-art Constrained Deep Adaptive Clustering (CDAC) model for question only. By comparison CDAC model for the question data only gives only up to 13pp performance boost over the naive baseline.

EACL 2023

Going beyond research datasets: Novel intent discovery in the industry setting

novel intent discovery

representation learning

clustering

findings / work in progress

### Welcome to the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL) 
Continuing its mission of expanding and involving the science community of all European countries, EACL had selected the Ukrainian community for the 16th EACL, which took place online due to the COVID pandemic. Unfortunately, the ongoing war made the organisation in Kyiv impossible. Considering the importance of physical interaction among researchers, especially after the restrictions imposed by the COVID pandemic, in addition to an online mode, the conference will be held in Dubrovnik, Croatia, from 2 to 6 of May, 2023. The original aim of strengthening the connection with the Ukrainian community will not change as our program will feature a dedicated session and a workshop to highlight work on Ukrainian language technologies. As the flagship European conference in the field of computational linguistics, EACL welcomes European and international researchers covering a broad spectrum of research areas that are concerned with computational approaches to natural language.

You need to register for the conference in order to access this site. Please visit https://2023.eacl.org/registration for more information.

As the flagship European conference in the field of computational linguistics, EACL welcomes European and international researchers covering a broad spectrum of research areas that are concerned with computational approaches to natural language.

The rapid development of large pretrained language models has revolutionized not only the field of Natural Language Generation (NLG) but also its evaluation. Inspired by the recent work of BARTScore: a metric leveraging the BART language model to evaluate the quality of generated text from various aspects, we introduce DATScore. DATScore uses data augmentation techniques to improve the evaluation of machine translation. Our main finding is that introducing data augmented translations of the source and reference texts is greatly helpful in evaluating the quality of the generated translation. We also propose two novel score averaging and term weighting strategies to improve the original score computing process of BARTScore. Experimental results on WMT show that DATScore correlates better with human meta-evaluations than the other recent state-of-the-art metrics, especially for low-resource languages. Ablation studies demonstrate the value added by our new scoring strategies. Moreover, we report in our extended experiments the performance of DATScore on 3 NLG tasks other than translation.

DATScore: Evaluating Translation with Data Augmented Translations

Humans tend to follow the Uniform Information Density (UID) principle by distributing information evenly in utterances. We study if decoding algorithms implicitly follow this UID principle, and under what conditions adherence to UID might be desirable for dialogue generation. We generate responses using different decoding algorithms with GPT-2 on the Persona-Chat dataset and collect human judgments on their quality using Amazon Mechanical Turk. We find that (i) surprisingly, model-generated responses follow the UID principle to a greater extent than human responses, and (ii) decoding algorithms that promote UID do not generate higher-quality responses. Instead, when we control for surprisal, non-uniformity of information density correlates with the quality of responses with very low/high surprisal. Our findings indicate that encouraging non-uniform responses is a potential solution to the ``likelihood trap'' problem (quality degradation in very high-likelihood text). Our dataset containing multiple candidate responses per dialog history along with human-annotated quality ratings is available at: https://huggingface.co/datasets/saranya132/dialog_uid_gpt2.

How do decoding algorithms distribute information in dialogue responses?

In order to reliably process natural language, NLP systems must generalize to the long tail of rare utterances. We propose a method to create challenging benchmarks that require generalizing to the tail of the distribution by re-splitting existing datasets. We create 'Likelihood Splits' where examples that are assigned lower likelihood by a pre-trained language model (LM) are placed in the test set, and more likely examples are in the training set. This simple approach can be customized to construct meaningful train-test splits for a wide range of tasks. Likelihood Splits surface more challenges than random splits: relative error rates of state-of-the-art models increase by 59% for semantic parsing on Spider, 93% for natural language inference on SNLI, and 33% for yes/no question answering on BoolQ, on our splits compared with the corresponding random splits. Moreover, Likelihood Splits create fairer benchmarks than adversarial filtering; when the LM used to create the splits is also employed as the task model, our splits do not unfairly penalize the LM.

Benchmarking Long-tail Generalization with Likelihood Splits

Multilingual pretraining approaches in Neural Machine Translation (NMT) have shown that training models to denoise synthetic code-switched data can yield impressive performance gains — owing to better multilingual semantic representations and transfer learning. However, they generated the synthetic code-switched data using non-contextual, one-to-one word translations obtained from lexicons - which can lead to significant noise in a variety of cases, including the poor handling of polysemes and multi-word expressions, violation of linguistic agreement and inability to scale to agglutinative languages. To overcome these limitations, we propose an approach called Contextual Code-Switching (CCS), where contextual, many-to-many word translations are generated using a `base' NMT model. We conduct experiments on 3 different language families - Romance, Uralic, and Indo-Aryan - and show significant improvements (by up to 5.5 spBLEU points) over the previous lexicon-based SOTA approaches. We also observe that small CCS models can perform comparably or better than massive models like mBART50 and mRASP2, depending on the size of data provided. We empirically analyse several key factors responsible for these - including context, many-to-many substitutions, code-switching language count etc. - and prove that they all contribute to enhanced pretraining of multilingual NMT models.

Exploring Enhanced Code-Switched Noising for Pretraining in Neural Machine Translation

Dialogue State Tracking (DST), a crucial component of task-oriented dialogue (ToD) systems, keeps track of all important information pertaining to dialogue history: filling slots with the most probable values throughout the conversation. Existing methods generally rely on a predefined set of values and struggle to generalise to previously unseen slots in new domains. To overcome these challenges, we propose a domain-agnostic extractive question answering (QA) approach with shared weights across domains. To disentangle the complex domain information in ToDs, we train our DST with a novel domain filtering strategy by excluding out-of-domain question samples. With an independent classifier that predicts the presence of multiple domains given the context, our model tackles DST by extracting spans in active domains. Empirical results demonstrate that our model can efficiently leverage domain-agnostic QA datasets by two-stage fine-tuning while being both domain-scalable and open vocabulary in DST. It shows strong transferability by achieving zero-shot domain-adaptation results on MultiWOZ 2.1 with an average JGA of 36.7%. It further achieves cross-lingual transfer with state-of-the-art zero-shot results, 66.2% JGA from English to German and 75.7% JGA from English to Italian on WOZ 2.0.

XQA-DST: Multi-Domain and Multi-Lingual Dialogue State Tracking

When upgrading neural models to a newer version, new errors that were not encountered in the legacy version can be introduced, known as regression errors. This inconsistent behavior during model upgrade often outweighs the benefits of accuracy gain and hinders the adoption of new models. To mitigate regression errors from model upgrade, distillation and ensemble have proven to be viable solutions without significant compromise in performance. Despite the progress, these approaches attained an incremental reduction in regression which is still far from achieving backward-compatible model upgrade. In this work, we propose a novel method, Gated Fusion, that promotes backward compatibility via learning to mix predictions between old and new models. Empirical results on two distinct model upgrade scenarios show that our method reduces the number of regression errors by 62% on average, outperforming the strongest baseline by an average of 25%.

Improving Prediction Backward-Compatiblility in NLP Model Upgrade with Gated Fusion

Given a sentence "Abby told Brittney that she upset Courtney", one would struggle to understand who "she" refers to, and ask for clarification. However, if the word "upset" were replaced with "hugged", "she" unambiguously refers to Abby. We study if modern coreference resolution models are sensitive to such pronominal ambiguity. To this end, we construct AmbiCoref, a diagnostic corpus of minimal sentence pairs with ambiguous and unambiguous referents. Our examples generalize psycholinguistic studies of human perception of ambiguity around particular arrangements of verbs and their arguments. Analysis shows that (1) humans are less sure of referents in ambiguous AmbiCoref examples than unambiguous ones, and (2) most coreference models show little difference in output between ambiguous and unambiguous pairs. We release AmbiCoref as a diagnostic corpus for testing whether models treat ambiguity similarly to humans.

AmbiCoref: Evaluating Human and Model Sensitivity to Ambiguous Coreference

Unsupervised out-of-domain (OOD) detection is a task aimed at discriminating whether given samples are from the in-domain or not, without the categorical labels of in-domain instances. Unlike supervised OOD, as there are no labels for training a classifier, previous works on unsupervised OOD detection adopted the one-class classification (OCC) approach, assuming that the training samples come from a single domain. However, in-domain instances in many real world applications can have a heterogeneous distribution (i.e., across multiple domains or multiple classes). In this case, OCC methods have difficulty in reflecting the categorical information of the domain properly. To tackle this issue, we propose a two-stage framework that leverages the latent categorical information to improve representation learning for textual OOD detection. In the first stage, we train a transformer-based sentence encoder for pseudo labeling by contrastive loss and cluster loss. The second stage is pseudo label learning in which the model is re-trained with pseudo-labels obtained in the first stage. The empirical results on the three datasets show that our two-stage framework significantly outperforms baseline models in more challenging scenarios.

Improving Unsupervised Out-of-domain detection through Pseudo Labeling and Learning

Recently introduced instruction-paradigm empowers non-expert users to leverage NLP resources by defining a new task in natural language. Instruction-tuned models have significantly outperformed multitask learning models (without instruction); however they are far from state-of-the-art task-specific models. Conventional approaches to improve model performance via creating datasets with large number of task instances or architectural changes in the model may not be feasible for non-expert users. However, they can write alternate instructions to represent an instruction task. Is Instruction-augmentation helpful? We augment a subset of tasks in the expanded version of NATURAL INSTRUCTIONS with additional instructions and find that it significantly improves model performance (up to 35%), especially in the low-data regime. Our results indicate that an additional instruction can be equivalent to ~200 data samples on average across tasks.

How Many Data Samples is an Additional Instruction Worth?

Adversarial attack aims to perturb input sequences and mislead a trained model for false predictions. To enhance the model robustness, defensing methods are accordingly employed by either data augmentation (involving adversarial samples) or model enhancement (modifying the training loss and/or model architecture). In contrast to previous work, this paper revisits the masked language modeling (MLM) and presents a simple yet efficient algorithm against adversarial attacks, termed [MASK] insertion for defensing (MI4D). Specifically, MI4D simply inserts [MASK] tokens to input sequences during training and inference, maximizing the intersection of the new convex hull (MI4D creates) with the original one (the clean input forms). As neither additional adversarial samples nor the model modification is required, MI4D is as computationally efficient as traditional fine-tuning. Comprehensive experiments have been conducted using three benchmark datasets and four attacking methods. MI4D yields a significant improvement (on average) of the accuracy between 3.2 and 11.1 absolute points when compared with six state-of-the-art defensing baselines.

Premium content

Going beyond research datasets: Novel intent discovery in the industry setting

Downloads

Next from EACL 2023

DATScore: Evaluating Translation with Data Augmented Translations

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES