Analyse du discours | EPFL Graph Search

L’analyse de discours est un champ des sciences humaines et sociales dont l'objet est le discours et qui implique une considération méthodologique. L’analyse de discours est une approche multidisciplinaire, qualitative et quantitative, qui étudie le contexte et le contenu des discours oraux ou écrits. L’analyse de discours est une approche multidisciplinaire qui s'est développée en France, en Grande-Bretagne et aux États-Unis à partir des années 1960. Elle emprunte de nombreux concepts aux champs de la sociologie, de la philosophie, de la psychologie, de l’informatique, des sciences de la communication, de la linguistique et de la statistique textuelle ou de l’histoire. Elle s'applique à des objets aussi variés que le discours politique, religieux, scientifique, artistique. Contrairement à l’analyse de contenu, dans sa définition traditionnelle, l’analyse de discours s’intéresse aux concepts, à la linguistique et à l’organisation narrative des discours oraux et écrits qu’elle étudie. On dit que l’analyse de discours est une approche socio-sémantique. Elle prend en compte le contexte de l’énonciation, les caractéristiques des locuteurs ainsi que les caractéristiques sémantiques de l’énoncé. L’analyse de discours considère le fonctionnement linguistique des discours dans lesquels elle voit une réalité à analyser. L'analyse de discours emprunte généralement un bagage théorique et analytique en provenance de l'approche structurale ou post-structurale. L’analyse de discours étudie ce que Louis Althusser appelle les « formations imaginaires ». Ce sont les marques de subjectivité du locuteur, les figures rhétoriques (conscientes ou inconscientes), la grammaire et ses formes diverses. Elle étudie aussi ce que Mikhaïl Bakhtine appelle l’« intertextualité », c’est-à-dire la relation entre les textes ou la manière dont les textes réagissent entre eux. À l'instar de Michel Foucault, l'analyse du discours postule le discours oral ou écrit comme un univers dans lequel s'expriment des contraintes.

Natural Language Processing (NLP) driven categorisation and detection of discourse in historical US patents

Jérôme Baudry, Nicolas Christophe Chachereau, Bhargav Srinivasa Desikan, Prakhar Gupta

Patents have traditionally been used in the history of technology as an indication of the thinking process of the inventors, of the challenges or “reverse salients” they faced, or of the social groups influencing the construction of technology. More recently, historians of science and technology also read them to interpret the way people described technology and how the specific inscriptions of inventions mattered for the justification and operation of the patent system. The digitization of historical patents opens up unique opportunities to assess the feasibility of unsupervised machine learning and natural language methods for such explorations. In this project, we analyze over a million US historical patents from 1830-1930 using a variety of text-based methods, with two major aims: 1) categorizing patents into coherent technical categories, 2) identifying discourses of safety, reflexivity, and environmental concern in technological innovation. We use both frequency-based and context-based methods, and find that a bag of words-based methods such as TDF-IDF and topic modeling do not perform well on semantic categorization due to the linguistic peculiarities of patent specifications. This suggests that a successful approach to categorizing patents would require contextual semantic representations such as Transformers-based methods (e.g. BERT), or static embedding based methods (e.g Word2Vec, Doc2Vec) which have relatively low computational costs but less expressive in some scenarios. We run early experiments using these methods and find that word embedding models are effective in learning semantics from the descriptions of the patents. In this poster, we will describe our early results, as well as exploratory data analysis on this massive historical patents dataset.

2022

Discourse Phenomena in Machine Translation

Lesly Sadiht Miculicich Werlen

Machine Translation (MT) has made considerable progress in the past two decades, particularly after the introduction of neural network models (NMT). During this time, the research community has mostly focused on modeling and evaluating MT systems at the sentence level. MT models learn to translate from large amounts of parallel sentences in different languages. The focus on sentences brings a practical simplification for the task that favors efficiency but has the disadvantage of missing relevant contextual information. Several studies showed that the negative impact of this simplification is significant. One key point is that the discourse dependencies among distant words are ignored, resulting in a lack of coherence and cohesion in the text. The main objective of this thesis is to improve MT by including discourse-level constraints. In particular, we focus on the translation of the entity mentions. We summarize our contributions in four points. First, we define the evaluation process to assess entity translations (i.e., nouns and pronouns) and propose an automatic metric to measure this phenomenon. Second, we perform a proof-of-concept and analyze how effective it is to include entity coreference resolution (CR) in translation. We conclude that CR significantly helps pronoun translation and boosts the whole translation quality according to human judgment. Third, we focus on the discourse connections at the sentence level. We propose enhancing the sequential model to infer long-term connections by incorporating a âself-attentionâ mechanism. This mechanism gives direct and selective access to the context. Experiments in different language pairs show that our method outperforms various baselines, and the analysis confirms that the model emphasizes a broader context and captures syntactic-like structures. Fourth, we formulate the problem of document-level NMT and model inter-sentential connections among words with a hierarchical attention mechanism. Experiments on multiple data sets show significant improvement over two strong baselines and conclude that the source and target sidesâ contexts are mutually complementary. This set of results confirms that discourse significantly enhances translation quality, verifying our main thesis objective. Our secondary objective is to improve the CR task by modeling the underlying connections among entities at the document-level. This task is particularly challenging for current neural network models because it requires understanding and reasoning. First, we propose a method to detect entity mentions from partially annotated data. We then proposed to model coreference with a graph of entities encoded in a pre-trained language model as an internal structure. The experiments show that these methods outperform various baselines. CR has the potential to help MT and other text generation tasks by maintaining coherence between the entity mentions.

EPFL2021

Computational Methods for Audio-Visual Analysis of Emergent Leadership in Teams

Dairazalia Sanchez Cortes

Face-to-face interactions are part of everyday life, ranging from family to working in teams and to global communities. Social psychologists have long studied these interactions with the aim of understanding behavior, motivations, and emergence of interaction patterns. An organization is environment rich in daily interactions including structured periodic meetings, planning, brainstorming, negotiations, decision-making and informal gatherings and leaders play a key role in many of them. Leader face problems, propose solutions, make decisions, and often are the main source of inspiration of the employees. Identifying emergent leaders at early stages in organizations is a key issue in organizational behavioral research, and a new problem in social computing. The study of this phenomenon requires sensing of natural face-to-face interactions, automatic extraction of behavioral cues and reliable machine learning algorithms to identify emergent leaders. In this thesis we present a computational approach to analyze emergence of leadership in small groups using multimodal audio and visual features. In the computational framework, we first present an analysis on how an emergent leader is perceived in newly formed, small groups. We present the ELEA (Emergent LEadership Analysis) corpus collected with the aim of analyzing emergence of leaders. We propose to analyze emergent leaders, using a variety of nonverbal cues studied in social psychology and automatically extracted from audio and video streams. Our analysis address how the emergent leader is perceived by his/her peers in terms of speaking and visual active, and its relation with the most dominant person (including external observers’ perception). We then propose to investigate which individual nonverbal channel (or combination of features from different channels) provides better inferences of the emergent leader and related concepts using unsupervised and supervised methods. We use a supervised collective approach which adds relational information to the nonverbal cues and compare its performance, with the performance of supervised (non-collective) and unsupervised methods. We also propose to capture the social visual attention patterns from automatically extracted features from video, in order to analyze who receives or gives the largest amount of visual attention in the group. Finally, with the aim of understanding who receives the largest amount of visual attention while speaking and who has the highest dominance ratio (i.e., many occurrences of looking at others while speaking and few occurrences of looking at others while not speaking). We synchronize the audio and video streams to capture the speaking and attention activity patterns. We end our analysis exploring the impact of the verbal content (language style) in the interactions and its influence in the perception of emergent leaders. For the language style analysis, we propose to compute word categories extracted from manual transcriptions of the discussions as well as from automatically detected keywords. We propose to use a supervised method to obtain the relevant features, and to use only the top word categories to predict the emergent leader and related concepts in each group. We then propose to differentiate word categories, between highly context-related and context-free, to explore the feasibility to infer the emergent leader in a fully automatic approach from the context-free language style. This dissertation address an audio and visual analysis of the ubiquitous phenomenon of emergent leadership in a fully automatic computational approach from face-to-face interactions. The nonverbal behavioral analysis is inspired in previous works on social psychology in the context of emergent leadership and related concepts. The automatically extracted nonverbal features are modeled to feed state-of-the-art machine learning techniques in order to infer emergent leaders.

EPFL2013