In this article, we present a database of multimodal communication features extracted from debate speeches in the 2019 North American Universities Debate Championships (NAUDC). Feature sets were extracted from the visual (facial expression, gaze, and head pose), audio (PRAAT), and textual (word sentiment and linguistic category) modalities of raw video recordings of competitive collegiate debaters (N=716 6-minute recordings from 140 unique debaters). Each speech has an associated competition debate score (range: 67-96) from experienced judges as well as competitor demographic and per-round reflection surveys. We observe the fully multimodal model performs best in comparison to models trained on various compositions of individual modalities. We also find that the weights of some features (such as the expression ofjoyand the use of the word ”we”) change in direction between the aforementioned models. We use these results to highlight the value of a multimodal dataset for studying competitive, collegiate debate.

Recognizing humor from a video utterance requires understanding the verbal and non-verbal components as well as incorporating the appropriate context and external knowledge. In this paper, we propose Humor Knowledge enriched Transformer (HKT) that can capture the gist of a multimodal humorous expression by integrating the preceding context and external knowledge. We incorporate humor centric external knowledge into the model by capturing the ambiguity and sentiment present in the language. We encode all the language, acoustic, vision, and humor centric features separately using Transformer based encoders, followed by a cross attention layer to exchange information among them. Our model achieves 77.36% and 79.41% accuracy in humorous punchline detection on UR-FUNNY and MUStaRD datasets -- achieving a new state-of-the-art on both datasets with the margin of 4.93% and 2.94% respectively. Furthermore, we demonstrate that our model can capture interpretable, humor-inducing patterns from all modalities.

The combination of gestures, intonations, and textual content plays a key role in argument delivery.However, the current literature mostly considers textual content while assessing the quality of an argument, and is limited to datasets containing short sequences (18-48 words).In this paper, we study argument quality assessment in a multimodal context, and experiment on DBATES, a publicly available dataset of long debate videos.First, we propose a set of interpretable debate-centric features such as clarity, content variation, body movement cues, and pauses, inspired by theories of argumentation quality.Second, we design the Multimodal ARgument Quality assessor (MARQ) -a hierarchical neural network model that summarizes the multimodal signals on long sequences and enriches the multimodal embedding with debate-centric features.Our proposed MARQ model achieves an accuracy of 81.91% on the argument quality prediction task and outperforms established baseline models with an error rate reduction of 22.7%.Through ablation studies, we demonstrate the importance of multimodal cues in modeling argument quality.

Recent Transformer-based contextual word representations, including BERT and XLNet, have shown state-of-the-art performance in multiple disciplines within NLP. Fine-tuning the trained contextual models on task-specific datasets has been the key to achieving superior performance downstream. While fine-tuning these pre-trained models is straight-forward for lexical applications (applications with only language modality), it is not trivial for multimodal language (a growing area in NLP focused on modeling face-to-face communication). Pre-trained models don't have the necessary components to accept two extra modalities of vision and acoustic. In this paper, we proposed an attachment to BERT and XLNet called Multimodal Adaptation Gate (MAG). MAG allows BERT and XLNet to accept multimodal nonverbal data during fine-tuning. It does so by generating a shift to internal representation of BERT and XLNet; a shift that is conditioned on the visual and acoustic modalities. In our experiments, we study the commonly used CMU-MOSI and CMU-MOSEI datasets for multimodal sentiment analysis. Fine-tuning MAG-BERT and MAG-XLNet significantly boosts the sentiment analysis performance over previous baselines as well as language-only fine-tuning of BERT and XLNet. On the CMU-MOSI dataset, MAG-XLNet achieves human-level multimodal sentiment analysis performance for the first time in the NLP community.

In this paper, we introduce a framework to automatically distinguish between facial expression sequences associated with imagining vs. remembering while answering a question. Our experiment includes a baseline and relevant questioning technique in the context of deception with 220 participants (20 hours long). Baseline questioning includes participants being separately asked to remember and imagine an arbitrary experience. During the relevant questioning, participants were prompted to either lie or tell the truth about a certain task. We trained a neural network model on the baseline data and achieved an accuracy of 60% on classifying imagining vs. remembering, whereas human performance for this task is 51%. Relevant questioning included a set of questions, each of which became an independent response segment. Using a transfer learning approach, we use the pretrained model from the baseline to obtain an imagination probability score for each relevant response segment. We define this individual probability per response as the Imagination Index. We apply the imagination indices as a feature vector to classify the whole relevant section as truth vs. bluff with an accuracy of 70%, significantly outperforming the human performance of 52%.

This paper demonstrates that analyzing language patterns in light of their associated facial expressions elicits significant differences between deceptive and truthful communication. Facial Action Units (AU) were analyzed in video recordings (1.2M frames) of 151 dyadic conversations following an interrogation protocol, in which one of the participants is known to be either lying or telling the truth. Linguistic features were extracted from the transcripts using Linguistic Inquiry and Word Count (LIWC) dictionary. Our framework extracted facial-feature contexts automatically corresponding to high and low intensities of AU occurrences. This helped us dive deeper into answers corresponding to the video segments where the witnesses kept their eyes wide open (high intensity of AU05-upper lid raise). We found that in these segments, deceivers used significantly fewer `Seeing', `Perceptual' and `Cognitive' words and their answers were significantly shorter than truth-tellers.

Md Kamrul Hasan, Wasifur Rahman, AmirAli Bagher Zadeh, Jianyuan Zhong, Md Iftekhar Tanveer, Louis-Philippe Morency, Mohammed (Ehsan) Hoque. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

In this paper we introduce the Common Human Emotional Expression Set Encoder (CHEESE) framework for objectively determining which, if any, subsets of the facial action units associated with smiling are well represented by a small finite set of clusters according to an information theoretic metric. Smile-related AUs (6,7,10,12,14) in over 1.3M frames of facial expressions from 151 pairs of individuals playing a communication game involving deception were analyzed with CHEESE. The combination of AU6 (cheek raiser) and AU12 (lip corner puller) are shown to cluster well into five different types of expression. Liars showed high intensity AU6 and AU12 more often compared to honest speakers. Additionally, interrogators were found to express a higher frequency of low intensity AU6 with high intensity AU12 (i.e. polite smiles) when they were being lied to, suggesting that deception analysis should be done in consideration of both the message sender's and the receiver's facial expressions.