Harmonizing Code-mixed Conversations: Personality-assisted Code-mixed Response Generation in Dialogues

TL;DR

This work tackles code-mixed dialogue response generation by unsupervisedly extracting Big Five personality traits from conversations and integrating them into the dialogue context through a novel PA3 fusion mechanism that leverages context-aware and axial attention. The method is evaluated on the Hindi-English MaSaC dataset, showing consistent improvements in ROUGE, BLEU, and BERTScore over strong baselines, and is complemented by extensive human evaluation. Key contributions include the unsupervised personality identification pipeline, the PA3 fusion module, and thorough ablation studies that demonstrate the value of personality-aware axial fusion for coherent, person-centered Code-mixed responses. The results indicate significant practical impact for creating more natural and contextually appropriate code-mixed dialogue systems in multi-party settings.

Abstract

Code-mixing, the blending of multiple languages within a single conversation, introduces a distinctive challenge, particularly in the context of response generation. Capturing the intricacies of code-mixing proves to be a formidable task, given the wide-ranging variations influenced by individual speaking styles and cultural backgrounds. In this study, we explore response generation within code-mixed conversations. We introduce a novel approach centered on harnessing the Big Five personality traits acquired in an unsupervised manner from the conversations to bolster the performance of response generation. These inferred personality attributes are seamlessly woven into the fabric of the dialogue context, using a novel fusion mechanism, PA3. It uses an effective two-step attention formulation to fuse the dialogue and personality information. This fusion not only enhances the contextual relevance of generated responses but also elevates the overall performance of the model. Our experimental results, grounded in a dataset comprising of multi-party Hindi-English code-mix conversations, highlight the substantial advantages offered by personality-infused models over their conventional counterparts. This is evident in the increase observed in ROUGE and BLUE scores for the response generation task when the identified personality is seamlessly integrated into the dialogue context. Qualitative assessment for personality identification and response generation aligns well with our quantitative results.

Figures (8)

• Figure 1: Influence of personality on dialogue responses -- a neurotic speaker might respond negatively to the posed question, whereas an extrovert would likely provide a positive reply.
• Figure 2: Dataset description of MaSaC (Abbreviation: Dlgs: Dialogues, Utts: Utterances, sp: speakers, Ma: Maya, In: Indravardhan, Sa: Sahil, Mo: Monisha, Ro: Rosesh, Oth: Others).
• Figure 3: Outline of learning personality traits using the 'pseudo' task of response generation.
• Figure 4: Model architecture to fuse personality values with dialogue context. The PA3 module can be injected into any encoder-decoder architecture, and it takes as inputs the dialogue representation along with the personality trait definition representation. First, context-aware attention is used to learn personality-infused key and value pairs and axial attention is then used to combine query, key, and value vectors into one final representation.
• Figure 5: Distribution of the predicted personality traits assigned to different speakers (Abbr - Ma: Maya, In: Indravardhan, Sa: Sahil, Mo: Monisha, Ro: Rosesh, Oth: Others).