Due to the lack of publicly available resources, conversation summarization has received far less attention than text summarization. As the purpose of conversations is to exchange information between at least two interlocutors, key information about a certain topic is often scattered and spanned across multiple utterances and turns from different speakers. This phenomenon is more pronounced during spoken conversations, where speech characteristics such as backchanneling and false-starts might interrupt the topical flow. Moreover, topic diffusion and (intra-utterance) topic drift are also more common in human-to-human conversations. Such linguistic characteristics of dialogue topics make sentence-level extractive summarization approaches used in spoken documents ill-suited for summarizing conversations. Pointer-generator networks have effectively demonstrated its strength at integrating extractive and abstractive capabilities through neural modeling in text summarization. To the best of our knowledge, to date no one has adopted it for summarizing conversations. In this work, we propose a topic-aware architecture to exploit the inherent hierarchical structure in conversations to further adapt the pointer-generator model. Our approach significantly outperforms competitive baselines, achieves more efficient learning outcomes, and attains more robust performance.

Topic-Aware Pointer-Generator Networks for Summarizing Spoken Conversations

Efficient communication during crisis response situations is a major challenge for involved emergency responders. Lack of relevant information or too much irrelevant information hampers the emergency responders’ decision-making process, workflow and situational awareness. Despite efforts to better centralize relevant information during crisis response, a gap still exists between the information supply and information needs of responders. Our contribution to bridge the information gap is a software system that monitors communication and may send information to emergency responders that were not addressed in the initial communication. The system, Task-Adaptive Information Distribution (TAID) is capable of disseminating information in a timely manner and adapting itself to the fast-changing information needs in a crisis response environment. This TAID system was trained with practical examples for which information relevance is known. To assess relevance, TAID uses a built relevance model for crisis response using methods from machine learning. In TAID this was set up as a classification task in which input information and knowledge about (ir)relevance of information was used. The technical results of the built relevance models are promising. The TAID-effect on crisis response was measured with simulation experiments which show that the expected effect of the TAID system on crisis response is restricted to specific circumstances. In a number of cases the intervening of TAID improved the situational awareness of emergency responders and reduced the duration of the crisis with very low additional costs, which results from time taken due to reading additional messages.

https://pure.uva.nl/ws/files/2481922/157981_206511_prs_Netten_BIBLIO_1.pdf

Machine learning for relevance of information in crisis response

We introduce a novel topic segmentation approach that combines evidence of topic shifts from lexical cohesion with linguistic evidence such as syntactically distinct features of segment initial and final contributions. Our evaluation shows that this hybrid approach outperforms state-of-the-art algorithms even when applied to loosely structured, spontaneous dialogue. Further analysis reveals that using dialogue exchanges versus dialogue contributions improves topic segmentation quality.

/pdf/topic-segmentation-of-dialogue-3h6vbnvshu.pdf

Topic-Segmentation of Dialogue

A system for passively analyzing an aggregation of communication signals simultaneously passing through a network link performs spectral analysis on a trace of communication traffic through the network link. The spectral analysis identifies period components of the trace corresponding to different communication signals. A system for deterring the passive analysis of signals aggregates communication traffic through a network link, such that a plurality of signals simultaneously pass through the network link. The plurality of signals is regulated such that the data packets corresponding to the plurality of signals leave the network link at a common rate. A memory temporarily stores the data packets as they enter the network link, and a timer maintains a fixed time interval for releasing the data packets from the memory.

Method and system for passively analyzing communication data based on frequency analysis of encrypted data traffic, and method and system for deterring passive analysis of communication data

Topic-aware Pointer-Generator Networks for Summarizing Spoken Conversations

We study the problem of topic segmentation of manually transcribed speech in order to facilitate information extraction from dialogs. Our approach is based on a combination of multi-source knowledge modeled by hidden Markov models. We experiment with different combinations of linguistic-level cues on dialogs dealing with search and rescue missions. Results show the effectiveness of multi-source knowledge.

/pdf/topic-segmentation-a-first-stage-to-dialog-based-information-3qlrfrtz85.pdf

Topic Segmentation : A First Stage to Dialog-Based Information Extraction.

Dans cet article, nous decrivons la complexite du traitement automatique des conversations. En particulier, nous etudions la problematique de l'extraction d'information a partir des conversations et nous presentons le decoupage thematique comme un outil d'aide a l'extraction.

Découpage thématique des conversations : un outil d'aide à l'extraction

Nous presentons les resultats de l’approche statistique que nous avons developpee pour le reperage de mots informatifs a partir de textes oraux. Ce travail fait partie d’un projet lance par le departement de la defense canadienne pour le developpement d’un systeme d’extraction d’information dans le domaine de la Recherche et Sauvetage maritime (SAR). Il s’agit de trouver et annoter les mots pertinents avec des etiquettes semantiques qui sont les concepts d’une ontologie du domaine (SAR). Notre methode combine deux types d’information : les vecteurs de similarite generes grâce a l’ontologie du domaine et le dictionnaire-thesaurus Wordsmyth ; le contexte d’enonciation represente par le theme. L’evaluation est effectuee en comparant la sortie du systeme avec les reponses de formulaires d’extraction d’information predefinis. Les resultats obtenus sur les textes oraux sont comparables a ceux obtenus dans le cadre de MUC7 pour des textes ecrits.

Approche statistique pour le repérage de mots informatifs dans les textes oraux

Nous presentons une approche de decoupage thematique que nous utiliserons pour faciliter l’extraction d’information a partir de conversations telephoniques transcrites. Nous experimentons avec un modele de Markov cache utilisant des informations de differents niveaux linguistiques, des marques d’extra-grammaticalites et les entites nommees comme source additionnelle d’information. Nous comparons le modele obtenu avec notre modele de base utilisant uniquement les marques linguistiques et les extra-grammaticalites. Les resultats montrent l’efficacite de l’approche utilisant les entites nommees.

/pdf/segmentation-en-themes-de-conversations-telephoniques-1icoqlmqrh.pdf

Segmentation en thèmes de conversations téléphoniques : traitement en amont pour l’extraction d’information

Nous presentons les resultats d'une approche d'etiquetage semantique developpee pour le reperage de mots informatifs a partir de textes conversationnels. Ce travail entre dans le cadre du developpement d'un systeme d'extraction d'information dans le domaine de la recherche et sauvetage maritime. Il s'agit de detecter et d'annoter les mots pertinents avec des etiquettes semantiques correspondant aux concepts d'une ontologie du domaine. Notre methode combine une approche symbolique basee sur un automate a etats finis et une approche statistique exploitant deux types d'information: les vecteurs de scores de similarite et le contexte discursif represente par le theme. Le F-score obtenu sur des transcriptions manuelles de conversations telephoniques dans le domaine de la recherche et sauvetage maritime est de 82,2 %.

Narjès Boufaden

Papers

Topic Segmentation : A First Stage to Dialog-Based Information Extraction.

Découpage thématique des conversations : un outil d'aide à l'extraction

Approche statistique pour le repérage de mots informatifs dans les textes oraux

Segmentation en thèmes de conversations téléphoniques : traitement en amont pour l’extraction d’information

Repérage de mots informatifs dans les textes conversationnels