Search

The development and evaluation of Large Language Models (LLMs) has primarily focused on their task-solving capabilities, with recent models even surpassing human performance in some areas. However, this focus often neglects whether machine-generated language matches the human level of diversity, in terms of vocabulary choice, syntactic construction, and expression of meaning, raising questions about whether the fundamentals of language generation have been fully addressed. This paper emphasizes the importance of examining the preservation of human linguistic richness by language models, given the concerning surge in online content produced or aided by LLMs. We propose a comprehensive framework for evaluating LLMs from various linguistic diversity perspectives including lexical, syntactic, and semantic dimensions. Using this framework, we benchmark several state-of-the-art LLMs across all diversity dimensions, and conduct an in-depth case study for syntactic diversity. Finally, we analyze how different development and deployment choices impact the linguistic diversity of LLM outputs.

Conversational systems are now capable of producing impressive and generally relevant responses. However, we have no visibility nor control of the socio-emotional strategies behind state-of-the-art Large Language Models (LLMs), which poses a problem in terms of their transparency and thus their trustworthiness for critical applications. Another issue is that current automated metrics are not able to properly evaluate the quality of generated responses beyond the dataset's ground truth. In this paper, we propose a neural architecture that includes an intermediate step in planning socio-emotional strategies before response generation. We compare the performance of open-source baseline LLMs to the outputs of these same models augmented with our planning module. We also contrast the outputs obtained from automated metrics and evaluation results provided by human annotators. We describe a novel evaluation protocol that includes a coarse-grained consistency evaluation, as well as a finer-grained annotation of the responses on various social and emotional criteria. Our study shows that predicting a sequence of expected strategy labels and using this sequence to generate a response yields better results than a direct end-to-end generation scheme. It also highlights the divergences and the limits of current evaluation metrics for generated content. The code for the annotation platform and the annotated data are made publicly available for the evaluation of future models.

Designing emotionally intelligent conversational systems to provide comfort and advice to people experiencing distress is a compelling area of research. Recently, with advancements in large language models (LLMs), end-to-end dialogue agents without explicit strategy prediction steps have become prevalent. However, implicit strategy planning lacks transparency, and recent studies show that LLMs' inherent preference bias towards certain socio-emotional strategies hinders the delivery of high-quality emotional support. To address this challenge, we propose decoupling strategy prediction from language generation, and introduce a novel dialogue strategy prediction framework, EmoDynamiX, which models the discourse dynamics between user fine-grained emotions and system strategies using a heterogeneous graph for better performance and transparency. Experimental results on two ESC datasets show EmoDynamiX outperforms previous state-of-the-art methods with a significant margin (better proficiency and lower preference bias). Our approach also exhibits better transparency by allowing backtracing of decision making.

Storytelling is an integral part of human experience and plays a crucial role in social interactions. Thus, Automatic Story Evaluation (ASE) and Generation (ASG) could benefit society in multiple ways, but they are challenging tasks which require high-level human abilities such as creativity, reasoning and deep understanding. Meanwhile, Large Language Models (LLM) now achieve state-of-the-art performance on many NLP tasks. In this paper, we study whether LLMs can be used as substitutes for human annotators for ASE. We perform an extensive analysis of the correlations between LLM ratings, other automatic measures, and human annotations, and we explore the influence of prompting on the results and the explainability of LLM behaviour. Most notably, we find that LLMs outperform current automatic measures for system-level evaluation but still struggle at providing satisfactory explanations for their answers., Comment: TACL, pre-MIT Press publication version

When deriving contextualized word representations from language models, a decision needs to be made on how to obtain one for out-of-vocabulary (OOV) words that are segmented into subwords. What is the best way to represent these words with a single vector, and are these representations of worse quality than those of in-vocabulary words? We carry out an intrinsic evaluation of embeddings from different models on semantic similarity tasks involving OOV words. Our analysis reveals, among other interesting findings, that the quality of representations of words that are split is often, but not always, worse than that of the embeddings of known words. Their similarity values, however, must be interpreted with caution., Comment: Accepted to TACL

With the increasing amount of problematic peer reviews in top AI conferences, the community is urgently in need of automatic quality control measures. In this paper, we restrict our attention to substantiation -- one popular quality aspect indicating whether the claims in a review are sufficiently supported by evidence -- and provide a solution automatizing this evaluation process. To achieve this goal, we first formulate the problem as claim-evidence pair extraction in scientific peer reviews, and collect SubstanReview, the first annotated dataset for this task. SubstanReview consists of 550 reviews from NLP conferences annotated by domain experts. On the basis of this dataset, we train an argument mining system to automatically analyze the level of substantiation in peer reviews. We also perform data analysis on the SubstanReview dataset to obtain meaningful insights on peer reviewing quality in NLP conferences over recent years., Comment: Accepted to EMNLP 2023 Findings

This study investigates the consequences of training language models on synthetic data generated by their predecessors, an increasingly prevalent practice given the prominence of powerful generative models. Diverging from the usual emphasis on performance metrics, we focus on the impact of this training methodology on linguistic diversity, especially when conducted recursively over time. To assess this, we adapt and develop a set of novel metrics targeting lexical, syntactic, and semantic diversity, applying them in recursive finetuning experiments across various natural language generation tasks in English. Our findings reveal a consistent decrease in the diversity of the model outputs through successive iterations, especially remarkable for tasks demanding high levels of creativity. This trend underscores the potential risks of training language models on synthetic text, particularly concerning the preservation of linguistic richness. Our study highlights the need for careful consideration of the long-term effects of such training approaches on the linguistic capabilities of language models., Comment: Accepted to NAACL 2024 Findings

We introduce MAFALDA, a benchmark for fallacy classification that merges and unites previous fallacy datasets. It comes with a taxonomy that aligns, refines, and unifies existing classifications of fallacies. We further provide a manual annotation of a part of the dataset together with manual explanations for each annotation. We propose a new annotation scheme tailored for subjective NLP tasks, and a new evaluation method designed to handle subjectivity. We then evaluate several language models under a zero-shot learning setting and human performances on MAFALDA to assess their capability to detect and classify fallacies.

This paper explores the application of machine learning techniques to predict where hedging occurs in peer-tutoring interactions. The study uses a naturalistic face-to-face dataset annotated for natural language turns, conversational strategies, tutoring strategies, and nonverbal behaviours. These elements are processed into a vector representation of the previous turns, which serves as input to several machine learning models. Results show that embedding layers, that capture the semantic information of the previous turns, significantly improves the model's performance. Additionally, the study provides insights into the importance of various features, such as interpersonal rapport and nonverbal behaviours, in predicting hedges by using Shapley values for feature explanation. We discover that the eye gaze of both the tutor and the tutee has a significant impact on hedge prediction. We further validate this observation through a follow-up ablation study., Comment: In Proceedings of the 16th Annual Conference ub Discourse and Dialogue (SIGDIAL). Sept 11-15, Prague Czechia

Hedging is a strategy for softening the impact of a statement in conversation. In reducing the strength of an expression, it may help to avoid embarrassment (more technically, ``face threat'') to one's listener. For this reason, it is often found in contexts of instruction, such as tutoring. In this work, we develop a model of hedge generation based on i) fine-tuning state-of-the-art language models trained on human-human tutoring data, followed by ii) reranking to select the candidate that best matches the expected hedging strategy within a candidate pool using a hedge classifier. We apply this method to a natural peer-tutoring corpus containing a significant number of disfluencies, repetitions, and repairs. The results show that generation in this noisy environment is feasible with reranking. By conducting an error analysis for both approaches, we reveal the challenges faced by systems attempting to accomplish both social and task-oriented goals in conversation.

Hedges play an important role in the management of conversational interaction. In peer tutoring, they are notably used by tutors in dyads (pairs of interlocutors) experiencing low rapport to tone down the impact of instructions and negative feedback. Pursuing the objective of building a tutoring agent that manages rapport with students in order to improve learning, we used a multimodal peer-tutoring dataset to construct a computational framework for identifying hedges. We compared approaches relying on pre-trained resources with others that integrate insights from the social science literature. Our best performance involved a hybrid approach that outperforms the existing baseline while being easier to interpret. We employ a model explainability tool to explore the features that characterize hedges in peer-tutoring conversations, and we identify some novel features, and the benefits of such a hybrid model approach., Comment: Published in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL), 2022

Disfluencies (i.e. interruptions in the regular flow of speech), are ubiquitous to spoken discourse. Fillers ("uh", "um") are disfluencies that occur the most frequently compared to other kinds of disfluencies. Yet, to the best of our knowledge, there isn't a resource that brings together the research perspectives influencing Spoken Language Understanding (SLU) on these speech events. This aim of this article is to survey a breadth of perspectives in a holistic way; i.e. from considering underlying (psycho)linguistic theory, to their annotation and consideration in Automatic Speech Recognition (ASR) and SLU systems, to lastly, their study from a generation standpoint. This article aims to present the perspectives in an approachable way to the SLU and Conversational AI community, and discuss moving forward, what we believe are the trends and challenges in each area., Comment: \footnote{This article has been published in the journal "Traitement Automatique des Langues" 63(3): 37-62, 2022,@ATALA. The original manuscript is available on the web site www.atala.org}

The topic of summarization evaluation has recently attracted a surge of attention due to the rapid development of abstractive summarization systems. However, the formulation of the task is rather ambiguous, neither the linguistic nor the natural language processing community has succeeded in giving a mutually agreed-upon definition. Due to this lack of well-defined formulation, a large number of popular abstractive summarization datasets are constructed in a manner that neither guarantees validity nor meets one of the most essential criteria of summarization: factual consistency. In this paper, we address this issue by combining state-of-the-art factual consistency models to identify the problematic instances present in popular summarization datasets. We release SummFC, a filtered summarization dataset with improved factual consistency, and demonstrate that models trained on this dataset achieve improved performance in nearly all quality aspects. We argue that our dataset should become a valid benchmark for developing and evaluating summarization systems., Comment: Accepted to EMNLP 2022

Research on Automatic Story Generation (ASG) relies heavily on human and automatic evaluation. However, there is no consensus on which human evaluation criteria to use, and no analysis of how well automatic criteria correlate with them. In this paper, we propose to re-evaluate ASG evaluation. We introduce a set of 6 orthogonal and comprehensive human criteria, carefully motivated by the social sciences literature. We also present HANNA, an annotated dataset of 1,056 stories produced by 10 different ASG systems. HANNA allows us to quantitatively evaluate the correlations of 72 automatic metrics with human criteria. Our analysis highlights the weaknesses of current metrics for ASG and allows us to formulate practical recommendations for ASG evaluation., Comment: 43 pages, 38 figures. Proceedings of the 29th International Conference on Computational Linguistics (COLING 2022)

Image schema is a recurrent pattern of reasoning where one entity is mapped into another. Image schema is similar to conceptual metaphor and is also related to metaphoric gesture. Our main goal is to generate metaphoric gestures for an Embodied Conversational Agent. We propose a technique to learn the vector representation of image schemas. As far as we are aware of, this is the first work which addresses that problem. Our technique uses Ravenet et al's algorithm which we use to compute the image schemas from the text input and also BERT and SenseBERT which we use as the base word embedding technique to calculate the final vector representation of the image schema. Our representation learning technique works by clustering: word embedding vectors which belong to the same image schema should be relatively closer to each other, and thus form a cluster. With the image schemas representable as vectors, it also becomes possible to have a notion that some image schemas are closer or more similar to each other than to the others because the distance between the vectors is a proxy of the dissimilarity between the corresponding image schemas. Therefore, after obtaining the vector representation of the image schemas, we calculate the distances between those vectors. Based on these, we create visualizations to illustrate the relative distances between the different image schemas.

Some exciting new approaches to neural architectures for the analysis of conversation have been introduced over the past couple of years. These include neural architectures for detecting emotion, dialogue acts, and sentiment polarity. They take advantage of some of the key attributes of contemporary machine learning, such as recurrent neural networks with attention mechanisms and transformer-based approaches. However, while the architectures themselves are extremely promising, the phenomena they have been applied to to date are but a small part of what makes conversation engaging. In this paper we survey these neural architectures and what they have been applied to. On the basis of the social science literature, we then describe what we believe to be the most fundamental and definitional feature of conversation, which is its co-construction over time by two or more interlocutors. We discuss how neural architectures of the sort surveyed could profitably be applied to these more fundamental aspects of conversation, and what this buys us in terms of a better analysis of conversation and even, in the longer term, a better way of generating conversation for a conversational system.

Assessing the quality of natural language generation systems through human annotation is very expensive. Additionally, human annotation campaigns are time-consuming and include non-reusable human labour. In practice, researchers rely on automatic metrics as a proxy of quality. In the last decade, many string-based metrics (e.g., BLEU) have been introduced. However, such metrics usually rely on exact matches and thus, do not robustly handle synonyms. In this paper, we introduce InfoLM a family of untrained metrics that can be viewed as a string-based metric that addresses the aforementioned flaws thanks to a pre-trained masked language model. This family of metrics also makes use of information measures allowing the adaptation of InfoLM to various evaluation criteria. Using direct assessment, we demonstrate that InfoLM achieves statistically significant improvement and over $10$ points of correlation gains in many configurations on both summarization and data2text generation.

se of machine learning for automatic analysis of job interview videos has recently seen increased interest. Despite claims of fair output regarding sensitive information such as gender or ethnicity of the candidates, the current approaches rarely provide proof of unbiased decision-making, or that sensitive information is not used. Recently, adversarial methods have been proved to effectively remove sensitive information from the latent representation of neural networks. However, these methods rely on the use of explicitly labeled protected variables (e.g. gender), which cannot be collected in the context of recruiting in some countries (e.g. France). In this article, we propose a new adversarial approach to remove sensitive information from the latent representation of neural networks without the need to collect any sensitive variable. Using only a few frames of the interview, we train our model to not be able to find the face of the candidate related to the job interview in the inner layers of the model. This, in turn, allows us to remove relevant private information from these layers. Comparing our approach to a standard baseline on a public dataset with gender and ethnicity annotations, we show that it effectively removes sensitive information from the main network. Moreover, to the best of our knowledge, this is the first application of adversarial techniques for obtaining a multimodal fair representation in the context of video job interviews. In summary, our contributions aim at improving fairness of the upcoming automatic systems processing videos of job interviews for equality in job selection., Comment: published in ACII 2021

Sequence-to-sequence neural networks have been widely used in language-based applications as they have flexible capabilities to learn various language models. However, when seeking for the optimal language response through trained neural networks, current existing approaches such as beam-search decoder strategies are still not able reaching to promising performances. Instead of developing various decoder strategies based on a "regular sentence order" neural network (a trained model by outputting sentences from left-to-right order), we leveraged "reverse" order as additional language model (a trained model by outputting sentences from right-to-left order) which can provide different perspectives for the path finding problems. In this paper, we propose bidirectional strategies in searching paths by combining two networks (left-to-right and right-to-left language models) making a bidirectional beam search possible. Besides, our solution allows us using any similarity measure in our sentence selection criterion. Our approaches demonstrate better performance compared to the unidirectional beam search strategy.

Several recent studies on dyadic human-human interactions have been done on conversations without specific business objectives. However, many companies might benefit from studies dedicated to more precise environments such as after sales services or customer satisfaction surveys. In this work, we place ourselves in the scope of a live chat customer service in which we want to detect emotions and their evolution in the conversation flow. This context leads to multiple challenges that range from exploiting restricted, small and mostly unlabeled datasets to finding and adapting methods for such context.We tackle these challenges by using Few-Shot Learning while making the hypothesis it can serve conversational emotion classification for different languages and sparse labels. We contribute by proposing a variation of Prototypical Networks for sequence labeling in conversation that we name ProtoSeq. We test this method on two datasets with different languages: daily conversations in English and customer service chat conversations in French. When applied to emotion classification in conversations, our method proved to be competitive even when compared to other ones.

Multimodal sentiment analysis is a trending area of research, and the multimodal fusion is one of its most active topic. Acknowledging humans communicate through a variety of channels (i.e visual, acoustic, linguistic), multimodal systems aim at integrating different unimodal representations into a synthetic one. So far, a consequent effort has been made on developing complex architectures allowing the fusion of these modalities. However, such systems are mainly trained by minimising simple losses such as $L_1$ or cross-entropy. In this work, we investigate unexplored penalties and propose a set of new objectives that measure the dependency between modalities. We demonstrate that our new penalties lead to a consistent improvement (up to $4.3$ on accuracy) across a large variety of state-of-the-art models on two well-known sentiment analysis datasets: \texttt{CMU-MOSI} and \texttt{CMU-MOSEI}. Our method not only achieves a new SOTA on both datasets but also produces representations that are more robust to modality drops. Finally, a by-product of our methods includes a statistical network which can be used to interpret the high dimensional representations learnt by the model.

Spoken dialog systems need to be able to handle both multiple languages and multilinguality inside a conversation (\textit{e.g} in case of code-switching). In this work, we introduce new pretraining losses tailored to learn multilingual spoken dialog representations. The goal of these losses is to expose the model to code-switched language. To scale up training, we automatically build a pretraining corpus composed of multilingual conversations in five different languages (French, Italian, English, German and Spanish) from \texttt{OpenSubtitles}, a huge multilingual corpus composed of 24.3G tokens. We test the generic representations on \texttt{MIAM}, a new benchmark composed of five dialog act corpora on the same aforementioned languages as well as on two novel multilingual downstream tasks (\textit{i.e} multilingual mask utterance retrieval and multilingual inconsistency identification). Our experiments show that our new code switched-inspired losses achieve a better performance in both monolingual and multilingual settings.

A new metric \texttt{BaryScore} to evaluate text generation based on deep contextualized embeddings e.g., BERT, Roberta, ELMo) is introduced. This metric is motivated by a new framework relying on optimal transport tools, i.e., Wasserstein distance and barycenter. By modelling the layer output of deep contextualized embeddings as a probability distribution rather than by a vector embedding; this framework provides a natural way to aggregate the different outputs through the Wasserstein space topology. In addition, it provides theoretical grounds to our metric and offers an alternative to available solutions e.g., MoverScore and BertScore). Numerical evaluation is performed on four different tasks: machine translation, summarization, data2text generation and image captioning. Our results show that \texttt{BaryScore} outperforms other BERT based metrics and exhibits more consistent behaviour in particular for text summarization.

Learning disentangled representations of textual data is essential for many natural language tasks such as fair classification, style transfer and sentence generation, among others. The existent dominant approaches in the context of text data {either rely} on training an adversary (discriminator) that aims at making attribute values difficult to be inferred from the latent code {or rely on minimising variational bounds of the mutual information between latent code and the value attribute}. {However, the available methods suffer of the impossibility to provide a fine-grained control of the degree (or force) of disentanglement.} {In contrast to} {adversarial methods}, which are remarkably simple, although the adversary seems to be performing perfectly well during the training phase, after it is completed a fair amount of information about the undesired attribute still remains. This paper introduces a novel variational upper bound to the mutual information between an attribute and the latent code of an encoder. Our bound aims at controlling the approximation error via the Renyi's divergence, leading to both better disentangled representations and in particular, a precise control of the desirable degree of disentanglement {than state-of-the-art methods proposed for textual data}. Furthermore, it does not suffer from the degeneracy of other losses in multi-class scenarios. We show the superiority of this method on fair classification and on textual style transfer tasks. Additionally, we provide new insights illustrating various trade-offs in style transfer when attempting to learn disentangled representations and quality of the generated sentence.

A dialogue is successful when there is alignment between the speakers at different linguistic levels. In this work, we consider the dialogue occurring between interlocutors engaged in a collaborative learning task, where they are not only evaluated on how well they performed, but also on how much they learnt. The main contribution of this work is to propose new automatic measures to study alignment; focusing on verbal (lexical) alignment, and behavioral alignment (when an instruction given by one was followed with concrete actions by another). A second contribution of our work is to study how spontaneous speech phenomena are used in the process of alignment. Lastly, we make public the dataset to study alignment in educational dialogues. Our results show that all teams verbally and behaviourally align to some degree regardless of their performance and learning, and our measures capture that teams that did not succeed in the task were simply slower to collaborate. Thus we find that teams that performed better, were faster to align. Furthermore, our methodology captures a productive period that includes the time where the interlocutors came up with their best solutions. We also find that well-performing teams verbalise the marker "oh" more when they are behaviourally aligned, compared to other times in the dialogue; showing that this marker is an important cue in alignment. To the best of our knowledge, we are the first to study the role of "oh" as an information management marker in a behavioral context (i.e. in connection to actions taken in a physical environment), compared to only a verbal one. Our measures contribute to the research in the field of educational dialogue and the intersection between dialogue and collaborative learning research., Comment: * The authors contributed equally to this work. This article a preprint under review

While being an essential component of spoken language, fillers (e.g."um" or "uh") often remain overlooked in Spoken Language Understanding (SLU) tasks. We explore the possibility of representing them with deep contextualised embeddings, showing improvements on modelling spoken language and two downstream tasks - predicting a speaker's stance and expressed confidence., Comment: To appear in EMNLP 2020

Sequence labelling tasks like Dialog Act and Emotion/Sentiment identification are a key component of spoken dialog systems. In this work, we propose a new approach to learn generic representations adapted to spoken dialog, which we evaluate on a new benchmark we call Sequence labellIng evaLuatIon benChmark fOr spoken laNguagE benchmark (\texttt{SILICONE}). \texttt{SILICONE} is model-agnostic and contains 10 different datasets of various sizes. We obtain our representations with a hierarchical encoder based on transformer architectures, for which we extend two well-known pre-training objectives. Pre-training is performed on OpenSubtitles: a large corpus of spoken dialog containing over $2.3$ billion of tokens. We demonstrate how hierarchical encoders achieve competitive results with consistently fewer parameters compared to state-of-the-art models and we show their importance for both pre-training and fine-tuning.

Communicative gestures and speech acoustic are tightly linked. Our objective is to predict the timing of gestures according to the acoustic. That is, we want to predict when a certain gesture occurs. We develop a model based on a recurrent neural network with attention mechanism. The model is trained on a corpus of natural dyadic interaction where the speech acoustic and the gesture phases and types have been annotated. The input of the model is a sequence of speech acoustic and the output is a sequence of gesture classes. The classes we are using for the model output is based on a combination of gesture phases and gesture types. We use a sequence comparison technique to evaluate the model performance. We find that the model can predict better certain gesture classes than others. We also perform ablation studies which reveal that fundamental frequency is a relevant feature for gesture prediction task. In another sub-experiment, we find that including eyebrow movements as acting as beat gesture improves the performance. Besides, we also find that a model trained on the data of one given speaker also works for the other speaker of the same conversation. We also perform a subjective experiment to measure how respondents judge the naturalness, the time consistency, and the semantic consistency of the generated gesture timing of a virtual agent. Our respondents rate the output of our model favorably.

In this paper, we consider the detection of a decrease of engagement by users spontaneously interacting with a socially assistive robot in a public space. We first describe the UE-HRI dataset that collects spontaneous Human-Robot Interactions following the guidelines provided by the Affective Computing research community to collect data "in-the-wild". We then analyze the users' behaviors, focusing on proxemics, gaze, head motion, facial expressions and speech during interactions with the robot. Finally, we investigate the use of deep learning techniques (Recurrent and Deep Neural Networks) to detect user engagement decrease in realtime. The results of this work highlight, in particular, the relevance of taking into account the temporal dynamics of a user's behavior. Allowing 1 to 2 seconds as buffer delay improves the performance of taking a decision on user engagement.

The dominant approaches to text representation in natural language rely on learning embeddings on massive corpora which have convenient properties such as compositionality and distance preservation. In this paper, we develop a novel method to learn a heavy-tailed embedding with desirable regularity properties regarding the distributional tails, which allows to analyze the points far away from the distribution bulk using the framework of multivariate extreme value theory. In particular, a classifier dedicated to the tails of the proposed embedding is obtained which performance outperforms the baseline. This classifier exhibits a scale invariance property which we leverage by introducing a novel text generation method for label preserving dataset augmentation. Numerical experiments on synthetic and real text data demonstrate the relevance of the proposed framework and confirm that this method generates meaningful sentences with controllable attribute, e.g. positive or negative sentiment.

The task of predicting dialog acts (DA) based on conversational dialog is a key component in the development of conversational agents. Accurately predicting DAs requires a precise modeling of both the conversation and the global tag dependencies. We leverage seq2seq approaches widely adopted in Neural Machine Translation (NMT) to improve the modelling of tag sequentiality. Seq2seq models are known to learn complex global dependencies while currently proposed approaches using linear conditional random fields (CRF) only model local tag dependencies. In this work, we introduce a seq2seq model tailored for DA classification using: a hierarchical encoder, a novel guided attention mechanism and beam search applied to both training and inference. Compared to the state of the art our model does not require handcrafted features and is trained end-to-end. Furthermore, the proposed approach achieves an unmatched accuracy score of 85% on SwDA, and state-of-the-art accuracy score of 91.6% on MRDA., Comment: in French

The task of predicting dialog acts (DA) based on conversational dialog is a key component in the development of conversational agents. Accurately predicting DAs requires a precise modeling of both the conversation and the global tag dependencies. We leverage seq2seq approaches widely adopted in Neural Machine Translation (NMT) to improve the modelling of tag sequentiality. Seq2seq models are known to learn complex global dependencies while currently proposed approaches using linear conditional random fields (CRF) only model local tag dependencies. In this work, we introduce a seq2seq model tailored for DA classification using: a hierarchical encoder, a novel guided attention mechanism and beam search applied to both training and inference. Compared to the state of the art our model does not require handcrafted features and is trained end-to-end. Furthermore, the proposed approach achieves an unmatched accuracy score of 85% on SwDA, and state-of-the-art accuracy score of 91.6% on MRDA.

The impact of non verbal behaviour in a hiring decision remains an open question. Investigating this question is important, as it could provide a better understanding on how to train candidates for job interviews and make recruiters be aware of influential non verbal behaviour. This research has recently been accelerated due to the development of tools for the automatic analysis of social signals, and the emergence of machine learning methods. However, these studies are still mainly based on hand engineered features, which imposes a limit to the discovery of influential social signals. On the other side, deep learning methods are a promising tool to discover complex patterns without the necessity of feature engineering. In this paper, we focus on studying influential non verbal social signals in asynchronous job video interviews that are discovered by deep learning methods. We use a previously published deep learning system that aims at inferring the hirability of a candidate with regard to a sequence of interview questions. One particularity of this system is the use of attention mechanisms, which aim at identifying the relevant parts of an answer. Thus, information at a fine-grained temporal level could be extracted using global (at the interview level) annotations on hirability. While most of the deep learning systems use attention mechanisms to offer a quick visualization of slices when a rise of attention occurs, we perform an in-depth analysis to understand what happens during these moments. First, we propose a methodology to automatically extract slices where there is a rise of attention (attention slices). Second, we study the content of attention slices by comparing them with randomly sampled slices. Finally, we show that they bear significantly more information for hirability than randomly sampled slices., Comment: Accepted at 2019 8th International Conference on Affective Computing and Intelligent Interaction (ACII)

The task of predicting fine grained user opinion based on spontaneous spoken language is a key problem arising in the development of Computational Agents as well as in the development of social network based opinion miners. Unfortunately, gathering reliable data on which a model can be trained is notoriously difficult and existing works rely only on coarsely labeled opinions. In this work we aim at bridging the gap separating fine grained opinion models already developed for written language and coarse grained models developed for spontaneous multimodal opinion mining. We take advantage of the implicit hierarchical structure of opinions to build a joint fine and coarse grained opinion model that exploits different views of the opinion expression. The resulting model shares some properties with attention-based models and is shown to provide competitive results on a recently released multimodal fine grained annotated corpus., Comment: Accepted to 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP) and 9th International Joint Conference on Natural Language Processing (IJCNLP)

New technologies drastically change recruitment techniques. Some research projects aim at designing interactive systems that help candidates practice job interviews. Other studies aim at the automatic detection of social signals (e.g. smile, turn of speech, etc...) in videos of job interviews. These studies are limited with respect to the number of interviews they process, but also by the fact that they only analyze simulated job interviews (e.g. students pretending to apply for a fake position). Asynchronous video interviewing tools have become mature products on the human resources market, and thus, a popular step in the recruitment process. As part of a project to help recruiters, we collected a corpus of more than 7000 candidates having asynchronous video job interviews for real positions and recording videos of themselves answering a set of questions. We propose a new hierarchical attention model called HireNet that aims at predicting the hirability of the candidates as evaluated by recruiters. In HireNet, an interview is considered as a sequence of questions and answers containing salient socials signals. Two contextual sources of information are modeled in HireNet: the words contained in the question and in the job position. Our model achieves better F1-scores than previous approaches for each modality (verbal content, audio and video). Results from early and late multimodal fusion suggest that more sophisticated fusion schemes are needed to improve on the monomodal results. Finally, some examples of moments captured by the attention mechanisms suggest our model could potentially be used to help finding key moments in an asynchronous job interview., Comment: AAAI 2019

In this paper, we introduce a set of opinion annotations for the POM movie review dataset, composed of 1000 videos. The annotation campaign is motivated by the development of a hierarchical opinion prediction framework allowing one to predict the different components of the opinions (e.g. polarity and aspect) and to identify the corresponding textual spans. The resulting annotations have been gathered at two granularity levels: a coarse one (opinionated span) and a finer one (span of opinion components). We introduce specific categories in order to make the annotation of opinions easier for movie reviews. For example, some categories allow the discovery of user recommendation and preference in movie reviews. We provide a quantitative analysis of the annotations and report the inter-annotator agreement under the different levels of granularity. We provide thus the first set of ground-truth annotations which can be used for the task of fine-grained multimodal opinion prediction. We provide an analysis of the data gathered through an inter-annotator study and show that a linear structured predictor learns meaningful features even for the prediction of scarce labels. Both the annotations and the baseline system are made publicly available. https://github.com/eusip/POM/

In this paper, the main goal is to detect a movie reviewer's opinion using hidden conditional random fields. This model allows us to capture the dynamics of the reviewer's opinion in the transcripts of long unsegmented audio reviews that are analyzed by our system. High level linguistic features are computed at the level of inter-pausal segments. The features include syntactic features, a statistical word embedding model and subjectivity lexicons. The proposed system is evaluated on the ICT-MMMO corpus. We obtain a F1-score of 82\%, which is better than logistic regression and recurrent neural network approaches. We also offer a discussion that sheds some light on the capacity of our system to adapt the word embedding model learned from general written texts data to spoken movie reviews and thus model the dynamics of the opinion., Comment: Oral Interspeech 2017

Motivated by Supervised Opinion Analysis, we propose a novel framework devoted to Structured Output Learning with Abstention (SOLA). The structure prediction model is able to abstain from predicting some labels in the structured output at a cost chosen by the user in a flexible way. For that purpose, we decompose the problem into the learning of a pair of predictors, one devoted to structured abstention and the other, to structured output prediction. To compare fully labeled training data with predictions potentially containing abstentions, we define a wide class of asymmetric abstention-aware losses. Learning is achieved by surrogate regression in an appropriate feature space while prediction with abstention is performed by solving a new pre-image problem. Thus, SOLA extends recent ideas about Structured Output Prediction via surrogate problems and calibration theory and enjoys statistical guarantees on the resulting excess risk. Instantiated on a hierarchical abstention-aware loss, SOLA is shown to be relevant for fine-grained opinion mining and gives state-of-the-art results on this task. Moreover, the abstention-aware representations can be used to competitively predict user-review ratings based on a sentence-level opinion predictor.