Jekaterina Novikova

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

BigScience Workshop, Teven Le Scao, Angela Fan et al. · allen-ai, berkeley

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License.

Sebastian Gehrmann, Abhik Bhattacharjee, Abinaya Mahendiran et al. · amazon-science, cmu

Evaluation in machine learning is usually informed by past choices, for example which datasets or metrics to use. This standardization enables the comparison on equal footing using leaderboards, but the evaluation choices become sub-optimal as better alternatives arise. This problem is especially pertinent in natural language generation which requires ever-improving suites of datasets, metrics, and human evaluation to make definitive claims. To make following best model evaluation practices easier, we introduce GEMv2. The new version of the Generation, Evaluation, and Metrics Benchmark introduces a modular infrastructure for dataset, model, and metric developers to benefit from each others work. GEMv2 supports 40 documented datasets in 51 languages. Models for all datasets can be evaluated online and our interactive data card creation and rendering tools make it easier to add new datasets to the living benchmark.

Malikeh Ehghaghi, Frank Rudzicz, Jekaterina Novikova

A significant number of studies apply acoustic and linguistic characteristics of human speech as prominent markers of dementia and depression. However, studies on discriminating depression from dementia are rare. Co-morbid depression is frequent in dementia and these clinical conditions share many overlapping symptoms, but the ability to distinguish between depression and dementia is essential as depression is often curable. In this work, we investigate the ability of clustering approaches in distinguishing between depression and dementia from human speech. We introduce a novel aggregated dataset, which combines narrative speech data from multiple conditions, i.e., Alzheimer's disease, mild cognitive impairment, healthy control, and depression. We compare linear and non-linear clustering approaches and show that non-linear clustering techniques distinguish better between distinct disease clusters. Our interpretability analysis shows that the main differentiating symptoms between dementia and depression are acoustic abnormality, repetitiveness (or circularity) of speech, word finding difficulty, coherence impairment, and differences in lexical complexity and richness.

Mashrura Tasnim, Malikeh Ehghaghi, Brian Diep et al.

Mental distress like depression and anxiety contribute to the largest proportion of the global burden of diseases. Automated diagnosis systems of such disorders, empowered by recent innovations in Artificial Intelligence, can pave the way to reduce the sufferings of the affected individuals. Development of such systems requires information-rich and balanced corpora. In this work, we introduce a novel mental distress analysis audio dataset DEPAC, labeled based on established thresholds on depression and anxiety standard screening tools. This large dataset comprises multiple speech tasks per individual, as well as relevant demographic information. Alongside, we present a feature set consisting of hand-curated acoustic and linguistic features, which were found effective in identifying signs of mental illnesses in human speech. Finally, we justify the quality and effectiveness of our proposed audio corpus and feature set in predicting depression severity by comparing the performance of baseline machine learning models built on this dataset with baseline models trained on other well-known depression corpora.

Nikita Dhawan, Arnav Paruthi, Andrew Kim et al.

Predicting the effect of interventions with many possible variations, e.g., therapeutic content that affects mental health outcomes or an earnings call transcript that drives movement in share price, is useful across several domains. However, classical causal estimators tend to assume that all possible interventions are observed, which is infeasible when interventions vary widely, for instance, in the space of all text strings. We adapt a well-known approach of recasting causal inference as a learning problem, to address high-dimensional treatment spaces. Specifically, under standard assumptions like no unobserved confounding, we show that causal error decomposes into a series of moment-balancing errors of increasing order, and design objectives that directly improve causal estimation. We also show how to project the effect of a high-dimensional treatment onto lower-dimensional treatment attributes, which allows a single model to answer several causal questions without additional attribute-specific training. We empirically evaluate our estimators in settings with high-dimensional continuous, discrete, and text treatments, the last of which used a semi-synthetic dataset of Amazon Reviews. Our experiments demonstrate the benefit of higher-order balance error optimization and competitive performance of projected causal estimates with attribute-specific estimators.

Brian Diep, Marija Stanojevic, Jekaterina Novikova

Traditional screening practices for anxiety and depression pose an impediment to monitoring and treating these conditions effectively. However, recent advances in NLP and speech modelling allow textual, acoustic, and hand-crafted language-based features to jointly form the basis of future mental health screening and condition detection. Speech is a rich and readily available source of insight into an individual's cognitive state and by leveraging different aspects of speech, we can develop new digital biomarkers for depression and anxiety. To this end, we propose a multi-modal system for the screening of depression and anxiety from self-administered speech tasks. The proposed model integrates deep-learned features from audio and text, as well as hand-crafted features that are informed by clinically-validated domain knowledge. We find that augmenting hand-crafted features with deep-learned features improves our overall classification F1 score comparing to a baseline of hand-crafted features alone from 0.58 to 0.63 for depression and from 0.54 to 0.57 for anxiety. The findings of our work suggest that speech-based biomarkers for depression and anxiety hold significant promise in the future of digital health.

Jekaterina Novikova, Ksenia Shkaruta

Models that accurately detect depression from text are important tools for addressing the post-pandemic mental health crisis. BERT-based classifiers' promising performance and the off-the-shelf availability make them great candidates for this task. However, these models are known to suffer from performance inconsistencies and poor generalization. In this paper, we introduce the DECK (DEpression ChecKlist), depression-specific model behavioural tests that allow better interpretability and improve generalizability of BERT classifiers in depression domain. We create 23 tests to evaluate BERT, RoBERTa and ALBERT depression classifiers on three datasets, two Twitter-based and one clinical interview-based. Our evaluation shows that these models: 1) are robust to certain gender-sensitive variations in text; 2) rely on the important depressive language marker of the increased use of first person pronouns; 3) fail to detect some other depression symptoms like suicidal ideation. We also demonstrate that DECK tests can be used to incorporate symptom-specific information in the training data and consistently improve generalizability of all three BERT models, with an out-of-distribution F1-score increase of up to 53.93%.

Malikeh Ehghaghi, Marija Stanojevic, Ali Akram et al.

Detecting duplicate patient participation in clinical trials is a major challenge because repeated patients can undermine the credibility and accuracy of the trial's findings and result in significant health and financial risks. Developing accurate automated speaker verification (ASV) models is crucial to verify the identity of enrolled individuals and remove duplicates, but the size and quality of data influence ASV performance. However, there has been limited investigation into the factors that can affect ASV capabilities in clinical environments. In this paper, we bridge the gap by conducting analysis of how participant demographic characteristics, audio quality criteria, and severity level of Alzheimer's disease (AD) impact the performance of ASV utilizing a dataset of speech recordings from 659 participants with varying levels of AD, obtained through multiple speech tasks. Our results indicate that ASV performance: 1) is slightly better on male speakers than on female speakers; 2) degrades for individuals who are above 70 years old; 3) is comparatively better for non-native English speakers than for native English speakers; 4) is negatively affected by clinician interference, noisy background, and unclear participant speech; 5) tends to decrease with an increase in the severity level of AD. Our study finds that voice biometrics raise fairness concerns as certain subgroups exhibit different ASV performances owing to their inherent voice characteristics. Moreover, the performance of ASV is influenced by the quality of speech recordings, which underscores the importance of improving the data collection settings in clinical trials.

Angelika Romanou, Negar Foroutan, Anna Sotnikova et al.

The performance differential of large language models (LLM) between languages hinders their effective deployment in many regions, inhibiting the potential economic and societal value of generative AI tools in many communities. However, the development of functional LLMs in many languages (\ie, multilingual LLMs) is bottlenecked by the lack of high-quality evaluation resources in languages other than English. Moreover, current practices in multilingual benchmark construction often translate English resources, ignoring the regional and cultural knowledge of the environments in which multilingual systems would be used. In this work, we construct an evaluation suite of 197,243 QA pairs from local exam sources to measure the capabilities of multilingual LLMs in a variety of regional contexts. Our novel resource, INCLUDE, is a comprehensive knowledge- and reasoning-centric benchmark across 44 written languages that evaluates multilingual LLMs for performance in the actual language environments where they would be deployed.

Jekaterina Novikova, Carol Anderson, Borhane Blili-Hamelin et al.

The hallmark of effective language use lies in consistency: expressing similar meanings in similar contexts and avoiding contradictions. While human communication naturally demonstrates this principle, state-of-the-art language models (LMs) struggle to maintain reliable consistency across task- and domain-specific applications. Here we examine the landscape of consistency research in LMs, analyze current approaches to measure aspects of consistency, and identify critical research gaps. Our findings point to an urgent need for quality benchmarks to measure and interdisciplinary approaches to ensure consistency while preserving utility.

Israfel Salazar, Manuel Fernández Burda, Shayekh Bin Islam et al. · mila

The evaluation of vision-language models (VLMs) has mainly relied on English-language benchmarks, leaving significant gaps in both multilingual and multicultural coverage. While multilingual benchmarks have expanded, both in size and languages, many rely on translations of English datasets, failing to capture cultural nuances. In this work, we propose Kaleidoscope, as the most comprehensive exam benchmark to date for the multilingual evaluation of vision-language models. Kaleidoscope is a large-scale, in-language multimodal benchmark designed to evaluate VLMs across diverse languages and visual inputs. Kaleidoscope covers 18 languages and 14 different subjects, amounting to a total of 20,911 multiple-choice questions. Built through an open science collaboration with a diverse group of researchers worldwide, Kaleidoscope ensures linguistic and cultural authenticity. We evaluate top-performing multilingual vision-language models and find that they perform poorly on low-resource languages and in complex multimodal scenarios. Our results highlight the need for progress on culturally inclusive multimodal evaluation frameworks.

Mardiyyah Oduwole, Prince Mireku, Fatimo Adebanjo et al.

Multimodal AI research has overwhelmingly focused on high-resource languages, hindering the democratization of advancements in the field. To address this, we present AfriCaption, a comprehensive framework for multilingual image captioning in 20 African languages and our contributions are threefold: (i) a curated dataset built on Flickr8k, featuring semantically aligned captions generated via a context-aware selection and translation process; (ii) a dynamic, context-preserving pipeline that ensures ongoing quality through model ensembling and adaptive substitution; and (iii) the AfriCaption model, a 0.5B parameter vision-to-text architecture that integrates SigLIP and NLLB200 for caption generation across under-represented languages. This unified framework ensures ongoing data quality and establishes the first scalable image-captioning resource for under-represented African languages, laying the groundwork for truly inclusive multimodal AI.

Ali Akram, Marija Stanojevic, Malikeh Ehghaghi et al.

Due to the substantial number of clinicians, patients, and data collection environments involved in clinical trials, gathering data of superior quality poses a significant challenge. In clinical trials, patients are assessed based on their speech data to detect and monitor cognitive and mental health disorders. We propose using these speech recordings to verify the identities of enrolled patients and identify and exclude the individuals who try to enroll multiple times in the same trial. Since clinical studies are often conducted across different countries, creating a system that can perform speaker verification in diverse languages without additional development effort is imperative. We evaluate pre-trained TitaNet, ECAPA-TDNN, and SpeakerNet models by enrolling and testing with speech-impaired patients speaking English, German, Danish, Spanish, and Arabic languages. Our results demonstrate that tested models can effectively generalize to clinical speakers, with less than 2.7% EER for European Languages and 8.26% EER for Arabic. This represents a significant step in developing more versatile and efficient speaker verification systems for cognitive and mental health clinical trials that can be used across a wide range of languages and dialects, substantially reducing the effort required to develop speaker verification systems for multiple languages. We also evaluate how speech tasks and number of speakers involved in the trial influence the performance and show that the type of speech tasks impacts the model performance.

Jekaterina Novikova

Research related to automatically detecting Alzheimer's disease (AD) is important, given the high prevalence of AD and the high cost of traditional methods. Since AD significantly affects the acoustics of spontaneous speech, speech processing and machine learning (ML) provide promising techniques for reliably detecting AD. However, speech audio may be affected by different types of background noise and it is important to understand how the noise influences the accuracy of ML models detecting AD from speech. In this paper, we study the effect of fifteen types of environmental noise from five different categories on the performance of four ML models trained with three types of acoustic representations. We perform a thorough analysis showing how ML models and acoustic features are affected by different types of acoustic noise. We show that acoustic noise is not necessarily harmful - certain types of noise are beneficial for AD detection models and help increasing accuracy by up to 4.8\%. We provide recommendations on how to utilize acoustic noise in order to achieve the best performance results with the ML models deployed in real world.

Jekaterina Novikova

Understanding robustness and sensitivity of BERT models predicting Alzheimer's disease from text is important for both developing better classification models and for understanding their capabilities and limitations. In this paper, we analyze how a controlled amount of desired and undesired text alterations impacts performance of BERT. We show that BERT is robust to natural linguistic variations in text. On the other hand, we show that BERT is not sensitive to removing clinically important information from text.

Aparna Balagopalan, Jekaterina Novikova

Robust strategies for Alzheimer's disease (AD) detection are important, given the high prevalence of AD. In this paper, we study the performance and generalizability of three approaches for AD detection from speech on the recent ADReSSo challenge dataset: 1) using conventional acoustic features 2) using novel pre-trained acoustic embeddings 3) combining acoustic features and embeddings. We find that while feature-based approaches have a higher precision, classification approaches relying on pre-trained embeddings prove to have a higher, and more balanced cross-validated performance across multiple metrics of performance. Further, embedding-only approaches are more generalizable. Our best model outperforms the acoustic baseline in the challenge by 2.8%.

Aparna Balagopalan, Jekaterina Novikova

Fine-tuned Bidirectional Encoder Representations from Transformers (BERT)-based sequence classification models have proven to be effective for detecting Alzheimer's Disease (AD) from transcripts of human speech. However, previous research shows it is possible to improve BERT's performance on various tasks by augmenting the model with additional information. In this work, we use probing tasks as introspection techniques to identify linguistic information not well-represented in various layers of BERT, but important for the AD detection task. We supplement these linguistic features in which representations from BERT are found to be insufficient with hand-crafted features externally, and show that jointly fine-tuning BERT in combination with these features improves the performance of AD classification by upto 5\% over fine-tuned BERT alone.

Benjamin Eyre, Aparna Balagopalan, Jekaterina Novikova

Despite the widely reported success of embedding-based machine learning methods on natural language processing tasks, the use of more easily interpreted engineered features remains common in fields such as cognitive impairment (CI) detection. Manually engineering features from noisy text is time and resource consuming, and can potentially result in features that do not enhance model performance. To combat this, we describe a new approach to feature engineering that leverages sequential machine learning models and domain knowledge to predict which features help enhance performance. We provide a concrete example of this method on a standard data set of CI speech and demonstrate that CI classification accuracy improves by 2.3% over a strong baseline when using features produced by this method. This demonstration provides an ex-ample of how this method can be used to assist classification in fields where interpretability is important, such as health care.

Aparna Balagopalan, Benjamin Eyre, Frank Rudzicz et al.

Research related to automatically detecting Alzheimer's disease (AD) is important, given the high prevalence of AD and the high cost of traditional methods. Since AD significantly affects the content and acoustics of spontaneous speech, natural language processing and machine learning provide promising techniques for reliably detecting AD. We compare and contrast the performance of two such approaches for AD detection on the recent ADReSS challenge dataset: 1) using domain knowledge-based hand-crafted features that capture linguistic and acoustic phenomena, and 2) fine-tuning Bidirectional Encoder Representations from Transformer (BERT)-based sequence classification models. We also compare multiple feature-based regression models for a neuropsychological score task in the challenge. We observe that fine-tuned BERT models, given the relative importance of linguistics in cognitive impairment detection, outperform feature-based approaches on the AD detection task.

Aparna Balagopalan, Jekaterina Novikova, Matthew B. A. McDermott et al.

Multi-language speech datasets are scarce and often have small sample sizes in the medical domain. Robust transfer of linguistic features across languages could improve rates of early diagnosis and therapy for speakers of low-resource languages when detecting health conditions from speech. We utilize out-of-domain, unpaired, single-speaker, healthy speech data for training multiple Optimal Transport (OT) domain adaptation systems. We learn mappings from other languages to English and detect aphasia from linguistic characteristics of speech, and show that OT domain adaptation improves aphasia detection over unilingual baselines for French (6% increased F1) and Mandarin (5% increased F1). Further, we show that adding aphasic data to the domain adaptation system significantly increases performance for both French and Mandarin, increasing the F1 scores further (10% and 8% increase in F1 scores for French and Mandarin, respectively, over unilingual baselines).

Jekaterina Novikova, Aparna Balagopalan, Ksenia Shkaruta et al.

Understanding the vulnerability of linguistic features extracted from noisy text is important for both developing better health text classification models and for interpreting vulnerabilities of natural language models. In this paper, we investigate how generic language characteristics, such as syntax or the lexicon, are impacted by artificial text alterations. The vulnerability of features is analysed from two perspectives: (1) the level of feature value change, and (2) the level of change of feature predictive power as a result of text modifications. We show that lexical features are more sensitive to text modifications than syntactic ones. However, we also demonstrate that these smaller changes of syntactic features have a stronger influence on classification performance downstream, compared to the impact of changes to lexical features. Results are validated across three datasets representing different text-classification tasks, with different levels of lexical and syntactic complexity of both conversational and written language.

Yuchen Li, Frank Rudzicz, Jekaterina Novikova

We seek to improve the data efficiency of neural networks and present novel implementations of parameterized piece-wise polynomial activation functions. The parameters are the y-coordinates of n+1 Chebyshev nodes per hidden unit and Lagrangian interpolation between the nodes produces the polynomial on [-1, 1]. We show results for different methods of handling inputs outside [-1, 1] on synthetic datasets, finding significant improvements in capacity of expression and accuracy of interpolation in models that compute some form of linear extrapolation from either ends. We demonstrate competitive or state-of-the-art performance on the classification of images (MNIST and CIFAR-10) and minimally-correlated vectors (DementiaBank) when we replace ReLU or tanh with linearly extrapolated Chebyshev-Lagrange activations in deep residual architectures.

Aparna Balagopalan, Ksenia Shkaruta, Jekaterina Novikova

Automatic Speech Recognition (ASR) is a critical component of any fully-automated speech-based dementia detection model. However, despite years of speech recognition research, little is known about the impact of ASR accuracy on dementia detection. In this paper, we experiment with controlled amounts of artificially generated ASR errors and investigate their influence on dementia detection. We find that deletion errors affect detection performance the most, due to their impact on the features of syntactic complexity and discourse representation in speech. We show the trend to be generalisable across two different datasets for cognitive impairment detection. As a conclusion, we propose optimising the ASR to reflect a higher penalty for deletion errors in order to improve dementia detection performance.

Ondřej Dušek, Jekaterina Novikova, Verena Rieser

This paper provides a comprehensive analysis of the first shared task on End-to-End Natural Language Generation (NLG) and identifies avenues for future research based on the results. This shared task aimed to assess whether recent end-to-end NLG systems can generate more complex output by learning from datasets containing higher lexical richness, syntactic complexity and diverse discourse phenomena. Introducing novel automatic and human metrics, we compare 62 systems submitted by 17 institutions, covering a wide range of approaches, including machine learning architectures -- with the majority implementing sequence-to-sequence models (seq2seq) -- as well as systems based on grammatical rules and templates. Seq2seq-based systems have demonstrated a great potential for NLG in the challenge. We find that seq2seq systems generally score high in terms of word-overlap metrics and human evaluations of naturalness -- with the winning SLUG system (Juraska et al., 2018) being seq2seq-based. However, vanilla seq2seq models often fail to correctly express a given meaning representation if they lack a strong semantic control mechanism applied during decoding. Moreover, seq2seq models can be outperformed by hand-engineered systems in terms of overall quality, as well as complexity, length and diversity of outputs. This research has influenced, inspired and motivated a number of recent studies outwith the original competition, which we also summarise as part of this paper.

Aparna Balagopalan, Jekaterina Novikova, Frank Rudzicz et al.

Speech datasets for identifying Alzheimer's disease (AD) are generally restricted to participants performing a single task, e.g. describing an image shown to them. As a result, models trained on linguistic features derived from such datasets may not be generalizable across tasks. Building on prior work demonstrating that same-task data of healthy participants helps improve AD detection on a single-task dataset of pathological speech, we augment an AD-specific dataset consisting of subjects describing a picture with multi-task healthy data. We demonstrate that normative data from multiple speech-based tasks helps improve AD detection by up to 9%. Visualization of decision boundaries reveals that models trained on a combination of structured picture descriptions and unstructured conversational speech have the least out-of-task error and show the most potential to generalize to multiple tasks. We analyze the impact of age of the added samples and if they affect fairness in classification. We also provide explanations for a possible inductive bias effect across tasks using model-agnostic feature anchors. This work highlights the need for heterogeneous datasets for encoding changes in multiple facets of cognition and for developing a task-independent AD detection model.

Ondřej Dušek, Jekaterina Novikova, Verena Rieser

This paper summarises the experimental setup and results of the first shared task on end-to-end (E2E) natural language generation (NLG) in spoken dialogue systems. Recent end-to-end generation systems are promising since they reduce the need for data annotation. However, they are currently limited to small, delexicalised datasets. The E2E NLG shared task aims to assess whether these novel approaches can generate better-quality output by learning from a dataset containing higher lexical richness, syntactic complexity and diverse discourse phenomena. We compare 62 systems submitted by 17 institutions, covering a wide range of approaches, including machine learning architectures -- with the majority implementing sequence-to-sequence models (seq2seq) -- as well as systems based on grammatical rules and templates.

Zining Zhu, Jekaterina Novikova, Frank Rudzicz

Linguistic features have shown promising applications for detecting various cognitive impairments. To improve detection accuracies, increasing the amount of data or the number of linguistic features have been two applicable approaches. However, acquiring additional clinical data can be expensive, and hand-crafting features is burdensome. In this paper, we take a third approach, proposing Consensus Networks (CNs), a framework to classify after reaching agreements between modalities. We divide linguistic features into non-overlapping subsets according to their modalities, and let neural networks learn low-dimensional representations that agree with each other. These representations are passed into a classifier network. All neural networks are optimized iteratively. In this paper, we also present two methods that improve the performance of CNs. We then present ablation studies to illustrate the effectiveness of modality division. To understand further what happens in CNs, we visualize the representations during training. Overall, using all of the 413 linguistic features, our models significantly outperform traditional classifiers, which are used by the state-of-the-art papers.

Zining Zhu, Jekaterina Novikova, Frank Rudzicz

One of the most prevalent symptoms among the elderly population, dementia, can be detected by classifiers trained on linguistic features extracted from narrative transcripts. However, these linguistic features are impacted in a similar but different fashion by the normal aging process. Aging is therefore a confounding factor, whose effects have been hard for machine learning classifiers (especially deep neural network based models) to ignore. We show DNN models are capable of estimating ages based on linguistic features. Predicting dementia based on this aging bias could lead to potentially non-generalizable accuracies on clinical datasets, if not properly deconfounded. In this paper, we propose to address this deconfounding problem with fair representation learning. We build neural network classifiers that learn low-dimensional representations reflecting the impacts of dementia yet discarding the effects of age. To evaluate these classifiers, we specify a model-agnostic score $Δ_{eo}^{(N)}$ measuring how classifier results are deconfounded from age. Our best models compromise accuracy by only 2.56\% and 1.54\% on two clinical datasets compared to DNNs, and their $Δ_{eo}^{(2)}$ scores are better than statistical (residulization and inverse probability weight) adjustments.

Zining Zhu, Jekaterina Novikova, Frank Rudzicz

Deep learning has demonstrated abilities to learn complex structures, but they can be restricted by available data. Recently, Consensus Networks (CNs) were proposed to alleviate data sparsity by utilizing features from multiple modalities, but they too have been limited by the size of labeled data. In this paper, we extend CN to Transductive Consensus Networks (TCNs), suitable for semi-supervised learning. In TCNs, different modalities of input are compressed into latent representations, which we encourage to become indistinguishable during iterative adversarial training. To understand TCNs two mechanisms, consensus and classification, we put forward its three variants in ablation studies on these mechanisms. To further investigate TCN models, we treat the latent representations as probability distributions and measure their similarities as the negative relative Jensen-Shannon divergences. We show that a consensus state beneficial for classification desires a stable but imperfect similarity between the representations. Overall, TCNs outperform or align with the best benchmark algorithms given 20 to 200 labeled samples on the Bank Marketing and the DementiaBank datasets.

Jekaterina Novikova, Ondřej Dušek, Verena Rieser

Human evaluation for natural language generation (NLG) often suffers from inconsistent user ratings. While previous research tends to attribute this problem to individual user preferences, we show that the quality of human judgements can also be improved by experimental design. We present a novel rank-based magnitude estimation method (RankME), which combines the use of continuous scales and relative assessments. We show that RankME significantly improves the reliability and consistency of human ratings compared to traditional evaluation methods. In addition, we show that it is possible to evaluate NLG systems according to multiple, distinct criteria, which is important for error analysis. Finally, we demonstrate that RankME, in combination with Bayesian estimation of system quality, is a cost-effective alternative for ranking multiple NLG systems.

Ondřej Dušek, Jekaterina Novikova, Verena Rieser

Traditional automatic evaluation measures for natural language generation (NLG) use costly human-authored references to estimate the quality of a system output. In this paper, we propose a referenceless quality estimation (QE) approach based on recurrent neural networks, which predicts a quality score for a NLG system output by comparing it to the source meaning representation only. Our method outperforms traditional metrics and a constant baseline in most respects; we also show that synthetic data helps to increase correlation results by 21% compared to the base system. Our results are comparable to results obtained in similar QE tasks despite the more challenging setting.

Jekaterina Novikova, Ondřej Dušek, Amanda Cercas Curry et al.

The majority of NLG evaluation relies on automatic metrics, such as BLEU . In this paper, we motivate the need for novel, system- and data-independent automatic evaluation methods: We investigate a wide range of metrics, including state-of-the-art word-based and novel grammar-based ones, and demonstrate that they only weakly reflect human judgements of system outputs as generated by data-driven, end-to-end NLG. We also show that metric performance is data- and system-specific. Nevertheless, our results also suggest that automatic metrics perform reliably at system-level and can support system development by finding cases where a system performs poorly.

Jekaterina Novikova, Ondřej Dušek, Verena Rieser

We argue that there are currently two major bottlenecks to the commercial use of statistical machine learning approaches for natural language generation (NLG): (a) The lack of reliable automatic evaluation metrics for NLG, and (b) The scarcity of high quality in-domain corpora. We address the first problem by thoroughly analysing current evaluation metrics and motivating the need for a new, more reliable metric. The second problem is addressed by presenting a novel framework for developing and evaluating a high quality corpus for NLG training.

Jekaterina Novikova, Ondřej Dušek, Verena Rieser

This paper describes the E2E data, a new dataset for training end-to-end, data-driven natural language generation systems in the restaurant domain, which is ten times bigger than existing, frequently used datasets in this area. The E2E dataset poses new challenges: (1) its human reference texts show more lexical richness and syntactic variation, including discourse phenomena; (2) generating from this set requires content selection. As such, learning from this dataset promises more natural, varied and less template-like system utterances. We also establish a baseline on this dataset, which illustrates some of the difficulties associated with this data.

Jekaterina Novikova, Christian Dondrup, Ioannis Papaioannou et al.

Recognition of social signals, from human facial expressions or prosody of speech, is a popular research topic in human-robot interaction studies. There is also a long line of research in the spoken dialogue community that investigates user satisfaction in relation to dialogue characteristics. However, very little research relates a combination of multimodal social signals and language features detected during spoken face-to-face human-robot interaction to the resulting user perception of a robot. In this paper we show how different emotional facial expressions of human users, in combination with prosodic characteristics of human speech and features of human-robot dialogue, correlate with users' impressions of the robot after a conversation. We find that happiness in the user's recognised facial expression strongly correlates with likeability of a robot, while dialogue-related features (such as number of human turns or number of sentences per robot utterance) correlate with perceiving a robot as intelligent. In addition, we show that facial expression, emotional features, and prosody are better predictors of human ratings related to perceived robot likeability and anthropomorphism, while linguistic and non-linguistic features more often predict perceived robot intelligence and interpretability. As such, these characteristics may in future be used as an online reward signal for in-situ Reinforcement Learning based adaptive human-robot dialogue systems.

Jekaterina Novikova, Oliver Lemon, Verena Rieser

Recent advances in corpus-based Natural Language Generation (NLG) hold the promise of being easily portable across domains, but require costly training data, consisting of meaning representations (MRs) paired with Natural Language (NL) utterances. In this work, we propose a novel framework for crowdsourcing high quality NLG training data, using automatic quality control measures and evaluating different MRs with which to elicit data. We show that pictorial MRs result in better NL data being collected than logic-based MRs: utterances elicited by pictorial MRs are judged as significantly more natural, more informative, and better phrased, with a significant increase in average quality ratings (around 0.5 points on a 6-point scale), compared to using the logical MRs. As the MR becomes more complex, the benefits of pictorial stimuli increase. The collected data will be released as part of this submission.