Do Neural Dialog Systems Use the Conversation History Effectively? An Empirical Study

By Chinnadhurai Sankar, Sandeep Subramanian, C. Pal, A. P. Sarath Chandar, Yoshua Bengio. 2019

125 citations recorded by the source corpus.

Neural generative models have been become increasingly popular when building conversational agents. They offer flexibility, can be easily adapted to new domains, and require minimal domain engineering. A common criticism of these systems is that they seldom understand or use the available dialog history effectively. In this paper, we take an empirical approach to understanding how these models use the available dialog history by studying the sensitivity of the models to artificially introduced unnatural changes or perturbations to their context at test time. We experiment with 10 different types of perturbations on 4 multi-turn dialog datasets and find that commonly used neural dialog architectures like recurrent and transformer-based seq2seq models are rarely sensitive to most perturbations such as missing or reordering utterances, shuffling words, etc. Also, by open-sourcing our code, we believe that it will serve as a useful diagnostic tool for evaluating dialog systems in the future.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Adversarial NLI: A New Benchmark for Natural Language Understanding (2019, neutral): “Similar observations were made in machine translation (Belinkov and Bisk, 2017) and dialogue (Sankar et al., 2019).”
Anticipating Safety Issues in E2E Conversational AI: Framework and Tooling (2021, background): “Most end-to-end systems, however, only use dialog history in a very limited fashion (Sankar et al., 2019).”
Dynabench: Rethinking Benchmarking in NLP (2021, background): “Dialogue models may ignore perturbations of dialogue history (Sankar et al., 2019).”
Assessing the Benchmarking Capacity of Machine Reading Comprehension Datasets (2019, methodology): “Our shufﬂe-based meth-ods ( s 6 to s 8 ) are inspired by existing analyses for other tasks (Khandelwal et al. 2018; Nie, Wang, and Bansal 2019; Sankar et al. 2019).”
Neural Generation Meets Real People: Towards Emotionally Engaging Mixed-Initiative Conversations (2020, background): “Given that neural models have been shown to make poor use of longer conversational history (Sankar et al., 2019), this truncation does not seem to be a limiting problem currently.”
What Context Features Can Transformer Language Models Use? (2021, background): “Similar effects are reported by Sankar et al. (2019) for neural dialogue models, and Li et al. (2016) describe an alternative procedure for ablating contexts.”
Mix-review: Alleviate Forgetting in the Pretrain-Finetune Framework for Neural Language Generation Models (2019, unknown): “The other is the utilization of contextual input (studied in Section 5.2): as shown by Sankar et al. (2019), the current open-domain dialogue models (without pre-training) are insensitive to contextual input, which gives rise to the generic response problem (Li et al., 2016).”
Reasoning with Transformer-based Models: Deep Learning, but Shallow Reasoning (2021, background): “Diﬀerent studies [Ettinger, 2020, Sankar et al., 2019, Pham et al., 2020, Gupta et al., 2021] have shown that BERT-like models are unperturbed by grammatically incorrect sentences: If presented with a sentence of randomly shuﬄed words, they will still reply correctly.”
Recent advances in deep learning based dialogue systems: a systematic survey (2021, methodology): “Open-domain systems apply RNNs as dialogue history encoders (Sankar et al., 2019; Du and Black, 2019; Ji et al., 2020; Chen et al., 2020b), among which retrieval-based systems model dialogue history and candidate responses together (Zhu et al., 2018; Tang et al., 2019; Feldman and El-Yaniv, 2019;…”
Analyzing the Forgetting Problem in Pretrain-Finetuning of Open-domain Dialogue Response Models (2021, methodology): “The other ability is the utilization of contextual input: as shown by (Sankar et al., 2019), the current open-domain dialogue models (without pretraining) are insensitive to contextual input, which gives rise to the generic response problem (Li et al.”
Hierarchical Reinforcement Learning for Open-Domain Dialog (2019, background): “Finally, models may have difficulty tracking long-term aspects of the conversation, and evidence has shown that they do not adequately condition on the conversation history in generating responses (Sankar et al. 2019).”
Sometimes We Want Ungrammatical Translations (2021, background): “For tasks like dialogue prediction that requires multiple input sentences, perturbation functions like reordering the conversation history have been adopted (Sankar et al., 2019).”
Conversations Are Not Flat: Modeling the Dynamic Information Flow across Dialogue Utterances (2021, unknown): “However, Sankar et al. (2019) demonstrate that ﬂat concatenation is likely to ignore the conversational dynamics across utterances in the dialogue history.”
Out of Order: How important is the sequential order of words in a sentence in Natural Language Understanding tasks? (2020, methodology): “Shuffling 1-grams is a common technique for analyzing word-order sensitivity (Sankar et al., 2019; Zanzotto et al., 2020).”
Context Dependent Semantic Parsing: A Survey (2020, background): “Thus, the quality of decoding depends tightly on the quality of contextual encoding, which is still a challenging problem (Sankar et al., 2019).”
A Compare Aggregate Transformer for Understanding Document-grounded Dialogue (2020, methodology): “Previous research [22] has shown that the last utterance is the most important guidance for the response generation in the multi-turn ar X iv :2 01 0.”
Lost in the Middle: How Language Models Use Long Contexts (2023, result): “The pioneering work of Khandelwal et al. (2018) showed that small LSTM language models make increasingly coarse use of longer-term context; Sankar et al. (2019) found similar results in dialogue models.”
Understanding and predicting user dissatisfaction in a neural generative chatbot (2021, background): “Though these models have known problems with factual correctness (Mielke et al., 2020), using dialogue history (Sankar et al., 2019), and bias (Dinan et al., 2020), they have nevertheless produced good written conversations when evaluated by crowdworkers or volunteers in carefullycontrolled…”