An Empirical Study of Self-Disclosure in Spoken Dialogue Systems

Abstract

Self-disclosure is a key social strategy employed in conversation to build relations and increase conversational depth. It has been heavily studied in psychology and linguistic literature, particularly for its ability to induce self-disclosure from the recipient, a phenomena known as reciprocity. However, we know little about how self-disclosure manifests in conversation with automated dialog systems, especially as any self-disclosure on the part of a dialog system is patently disingenuous. In this work, we run a large-scale quantitative analysis on the effect of selfdisclosure by analyzing interactions between real-world users and a spoken dialog system in the context of social conversation. We find that indicators of reciprocity occur even in human-machine dialog, with far-reaching implications for chatbots in a variety of domains including education, negotiation and social dialog.

Full Disclosure: How Smartphones Enhance Consumer Self-Disclosure

“…Consistent with prior work on the linguistic markers of self-disclosure, the examples show how texts judged to be self-disclosing tend to be accompanied by more extensive use of (1) first-person pronouns (e.g., “I,” “me”), (2) references to family and friends, and (3) words that convey emotionality—particularly negative emotions (Houghton and Joinson 2012; Okdie 2011). The presence of these common linguistic markers forms the foundation of algorithms designed to automatically detect depth of disclosure in online texts (Bak, Kim, and Oh 2012; Balani and De Choudhury 2015; Ravichander and Black 2018; Wang, Burke, and Kraut 2016).…”

“… 5 Exploratory analyses of other LIWC categories indicated that tweets written on smartphones (vs. PCs) also tended to use relatively more informal language (i.e., netspeak, nonfluencies, swear words). While a more informal writing style might also point to greater self-disclosure on smartphones, we focus on the four linguistic markers that have been validated in prior work (e.g., Ravichander and Black 2018). …”

Detecting Asks in Social Engineering Attacks: Impact of Linguistic and Structural Knowledge

“…To address this issue, topic models are used to produce focused responses augmenting existing neural based approaches to CAs (Dziri et al 2018). Targeted responses employ self-disclosure as a strategic approach for building an engaging conversation (Ravichander and Black 2018). For SE detection, topic models (Bhakta and Harris 2015) and NLP of conversations (Sawa et al 2016) are leveraged.…”

A Dynamic Speaker Model for Conversational Interactions

“…In both settings, topic decisions occurring in the first 5 user turns are not used for evaluations. Static User Embeddings: Motivated by the findings that most user characteristics can be inferred from initial interactions (Ravichander and Black, 2018), we derive a static user embedding vector for a conversation using the first 5 user turns and apply it for predicting topic decisions afterwards. Dynamic User Embeddings: Alternatively, we build a user embedding vector for user turn t using all previous user turns.…”