Analyzing Language Learned by an
Active Question Answering Agent
We analyze the language learned by an agent trained with reinforcement learning as a component of the ActiveQA system (Buck et al., 2017). In ActiveQA, question answering is framed as a reinforcement learning task in which an agent sits between the user and a black box question-answering system. The agent learns to reformulate the user’s questions to elicit the optimal answers. It probes the system with many versions of a question that are generated via a sequence-to-sequence question reformulation model, then aggregates the returned evidence to find the best answer. This process is an instance of machine-machine communication. The question reformulation model must adapt its language to increase the quality of the answers returned, matching the language of the question answering system. We find that the agent does not learn transformations that align with semantic intuitions but discovers through learning classical information retrieval techniques such as tf-idf re-weighting and stemming.
Analyzing Language Learned by an
Active Question Answering Agent
Christian Buck firstname.lastname@example.org Jannis Bulian email@example.com Massimiliano Ciaramita firstname.lastname@example.org Wojciech Gajewski email@example.com Andrea Gesmundo firstname.lastname@example.org Neil Houlsby email@example.com Wei Wang firstname.lastname@example.org Google
noticebox[b]31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.\end@float
Buck et al. (2017) propose a reinforcement learning framework for question answering, called active question answering (ActiveQA), that aims to improve answering by systematically perturbing input questions (cf. (Nogueira and Cho, 2017)). Figure 1 depicts the generic agent-environment framework. The agent (AQA) interacts with the environment (E) in order to answer a question (). The environment includes a question answering system (Q&A), and emits observations and rewards. A state at time is the sequence of observations and previous actions generated starting from : , where includes the question asked (), the corresponding answer returned by the QA system (), and possibly additional information such as features or auxiliary tasks. The agent includes an action scoring component (U), which produced and action by deciding whether to submit a new question to the environment or to return a final answer. Formally, , where is the set of all possible questions, and is the set of all possible answers. The agent relies on a question reformulation system (QR), that provides candidate follow up questions, and on an answer ranking system (AR), which scores the answers contained in . Each answer returned is assigned a reward. The objective is to maximize the expected reward over a set of questions.
Buck et al. (2017) present a simplified version of this system with three core components: a question reformulator, an off-the-shelf black box QA system, and a candidate answer selection model. The question reformulator is trained with policy gradient (Williams, 1992) to optimize the F1 score of the answers returned by the QA system to the question reformulations in place of the original question. The reformulator is implemented as a sequence-to-sequence model of the kind used for machine translation (Sutskever et al., 2014; Bahdanau et al., 2014). When generating question reformulations, the action-space is equal to the size of the vocabulary, typically sentence pieces.111https://github.com/google/sentencepiece Due to this large number of actions we warm start the reformulation policy with a monolingual sequence-to-sequence model that performs generic paraphrasing. This model is trained using the zero-shot translation technique (Johnson et al., 2016) on a large multilingual parallel corpus (Ziemski et al., 2016), followed by regular supervised learning on a smaller monolingual corpus of questions (Fader et al., 2013).
The reformulation and selection models form a trainable agent that seeks the best answers from the QA system. The reformulator proposes versions of the input question and passes them to the environment, which provides corresponding answers, . The selection model scores each triple and returns the top-scoring candidate.222For more details see (Buck et al., 2017).
Crucially, the agent may only query the environment with natural language questions. Thus, ActiveQA involves a machine-machine communication process inspired by the human-machine communication that takes place when users interact with digital services during information seeking tasks. For example, while searching for information on a search engine users tend to adopt a keyword-like ‘queryese’ style of questioning. The AQA agent proves effective at reformulating questions on SearchQA (Dunn et al., 2017), a large dataset of complex questions from the Jeopardy! game. For this task BiDAF is chosen for the environment (Seo et al., 2017), a deep network built for QA which has produced state-of-the-art results. Compared to a QA system that forms the environment using only the original questions, AQA outperforms this baseline by a wide margin, 11.4% absolute F1, thereby reducing the gap between machine (BiDAF) and human performance by 66%.
Here we perform a qualitative analysis of this communication process to better understand what kind of language the agent has learned. We find that while optimizing its reformulations to adapt to the language of the QA system, AQA diverges from well structured language in favour of less fluent, but more effective, classic information retrieval (IR) query operations. These include term re-weighting (tf-idf), expansion and morphological simplification/stemming. We hypothesize that the explanation of this behaviour is that current machine comprehension tasks primarily require ranking of short textual snippets, thus incentivizing relevance more than deep language understanding.
2 Analysis of the Agent’s Language
We analyze input questions and reformulations on the example development partition of the SearchQA dataset. Our goal is to gain insights on how the agent’s language evolves during training via policy gradient. It is important to note that in the SearchQA dataset the original Jeopardy! clues have been preprocessed by lower-casing and stop word removal. The resulting preprocessed clues that form the sources (inputs) for the sequence-to-sequence reformulation model resemble more keyword-based search queries than grammatical questions. For example, the clue Gandhi was deeply influenced by this count who wrote "War and Peace" is simplified to gandhi deeply influenced count wrote war peace.
2.1 The Language of SearchQA Questions
Figure 2 summarizes statistics of the questions and rewrites which may shed some light on how the language changes. The (preprocessed) SearchQA questions contain 9.6 words on average. They contain few repeated terms, computed as the mean term frequency (TF) per question. The average is 1.03, but for most of the queries TF is 1.0, i.e. no repetitions. We also compute the median document frequency (DF) per query, where a document is the context from which the answer is selected.333We use the median instead of the mean to reduce the influence of frequent outliers, such as commas, on the statistics. The mean DF is 460. DF gives a measure of how informative the question terms are.
2.2 The Language of the Base NMT Model
We first consider the top hypothesis generated by the pre-trained NMT reformulation system, before reinforcement learning (Base-NMT). This system is trained with full supervision, using a large multilingual and a small monolingual dataset. The Base-NMT rewrites differ greatly from their sources. They are shorter, 6.3 words on average, and have even fewer repeated terms (1.01). Interestingly, these reformulations are mostly syntactically well-formed questions. For example, the clue above becomes Who influenced count wrote war?.444More examples can be found in Appendix A. Base-NMT improves structural language quality by properly reinserting dropped function words and wh-phrases. We also verified the increased fluency by using a large language model and found that the Base-NMT rewrites are 50% more likely than the original questions. The bottom right hand plot in Figure 2 summarizes the language model distributions (LM WordLogP). The plot shows the average per-token language model negative log probabilities; a lower score indicates greater fluency. Although the distributions overlap to a great extent due to the large variance across questions, the differences in means are significant.
While more fluent, the Base-NMT rewrites involve rarer terms, as indicated by the decrease in DF. This is probably due to a domain mismatch between SearchQA and the NMT training corpus.
2.3 The Language of the AQA Agent
We next consider the top hypothesis generated by the AQA question reformulator (AQA-QR) after the policy gradient training. The AQA-QR rewrites are those whose corresponding answers are evaluated as AQA Top Hyp. in Buck et al. (2017). Note, these single rewrites alone outperform the original SearchQA queries by a small margin (+2% on test). We analyze the top hypothesis instead of the final output of the full AQA agent to avoid confounding effects from the answer selection step. These rewrites look different from both the Base-NMT and the SearchQA ones. For the example above AQA-QR’s top hypothesis is What is name gandhi gandhi influence wrote peace peace?. Surprisingly, 99.8% start with the prefix What is name. The second most frequent is What country is (81 times), followed by What is is (70) and What state (14). This is puzzling as it happens only for 9 Base-NMT rewrites, and never in the original SearchQA questions. We speculate it might be related to the fact that virtually all answers involve names, of named entities (Micronesia) or generic concepts (pizza). AQA-QR’s rewrites are visibly less fluent than both the SearchQA and the Base-MT counterparts. In terms of language model probability they are less likely than both SearchQA and Base-NMT.555To compute meaningful language model scores we remove the prefix “What is name” from all queries, because it artificially inflates the fluency measure, due to the high frequency unigrams and bigrams. However, they have more repeated terms (1.2 average TF), are significantly longer (11.9) than the Base-NMT initialization and contain more informative context terms (lower DF) than SearchQA questions.
Additionally, AQA-QR’s reformulations contain morphological variants in 12.5% of cases. The number of questions that contain multiple tokens with the same stem doubles from SearchQA to AQA-QR. Singular forms are preferred over plurals. Morphological simplification is useful because it increases the chance that a word variant in the question matches the context.
3 Conclusions: Rediscovering IR?
Recently, Lewis et al. (2017) trained chatbots that negotiate via language utterances in order to complete a task. They report that the agent’s language diverges from human language if there is no incentive for fluency in the reward function. Our findings seem related. The fact that the questions reformulated by AQA do not resemble natural language is not due to the keyword-like SearchQA input questions, because Base-NMT is capable of producing fluent questions from the same input.
AQA learns to re-weight terms by focusing on informative (lower DF) terms while increasing term frequency (TF) via duplication. At the same time it learns to modify surface forms in ways akin to stemming and morphological analysis. Some of the techniques seem to adapt also to the specific properties of current deep QA architectures such as character-based modelling and attention. Sometimes AQA learns to generate semantically nonsensical, novel, surface term variants; e.g., it might transform the adjective dense to densey. The only justification for this is that such forms can be still exploited by the character-based BiDAF question encoder. Finally, repetitions can directly increase the chances of alignment in the attention components.
We hypothesize that there is no incentive for the model to use human language due to the nature of the task. AQA learns to ask BiDAF questions by optimizing a language that increases the likelihood of BiDAF extracting the right answer. Jia and Liang (2017) argue that reading comprehension systems are not capable of significant language understanding and fail easily in adversarial settings. We suspect that current machine comprehension tasks involve mostly simple pattern matching and relevance modelling. As a consequence deep QA systems behave as sophisticated ranking systems trained to sort snippets of text from the context. As such, they resemble document retrieval systems which incentivizes the (re-)discovery of IR techniques that have been successful for decades (Baeza-Yates and Ribeiro-Neto, 1999).
- Baeza-Yates and Ribeiro-Neto  R. Baeza-Yates and B. Ribeiro-Neto. Modern Information Retrieval. Addison-Wesley Longman Publishing Co., Inc., 1999.
- Bahdanau et al.  D. Bahdanau, K. Cho, and Y. Bengio. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473, 2014.
- Buck et al.  C. Buck, J. Bulian, M. Ciaramita, W. Gajewski, A. Gesmundo, N. Houlsby, and W. Wang. Ask the right questions: Active query reformulation with reinforcement learning. https://arxiv.org/abs/1705.07830, 2017.
- Dunn et al.  M. Dunn, L. Sagun, M. Higgins, U. Guney, V. Cirik, and K. Cho. SearchQA: A New Q&A Dataset Augmented with Context from a Search Engine. https://arxiv.org/abs/1704.05179, 2017.
- Fader et al.  A. Fader, L. Zettlemoyer, and O. Etzioni. Paraphrase-Driven Learning for Open Question Answering. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, 2013.
- Jia and Liang  R. Jia and P. Liang. Adversarial examples for evaluating reading comprehension systems. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2011–2021, 2017.
- Johnson et al.  M. Johnson, M. Schuster, Q. V. Le, M. Krikun, Y. Wu, Z. Chen, N. Thorat, F. Viégas, M. Wattenberg, G. Corrado, M. Hughes, and J. Dean. Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation. https://arxiv.org/abs/1611.04558, 2016.
- Lewis et al.  M. Lewis, D. Yarats, Y. Dauphin, D. Parikh, and D. Batra. Deal or no deal? end-to-end learning of negotiation dialogues. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2433–2443, 2017.
- Nogueira and Cho  R. Nogueira and K. Cho. Task-oriented query reformulation with reinforcement learning. https://arxiv.org/abs/1704.04572v1, 2017.
- Seo et al.  M. Seo, A. Kembhavi, A. Farhadi, and H. Hajishirzi. Bidirectional Attention Flow for Machine Comprehension. In Proceedings of ICLR, 2017.
- Sutskever et al.  I. Sutskever, O. Vinyals, and Q. V. Le. Sequence to sequence learning with neural networks. In Proceedings of the 27th International Conference on Neural Information Processing Systems, pages 3104–3112, 2014.
- Williams  R. J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Mach. Learn., 8(3-4):229–256, 1992.
- Ziemski et al.  M. Ziemski, M. Junczys-Dowmunt, and B. Poliquen. The united nations parallel corpus v1.0. In Proceedings of Language Resources and Evaluation (LREC), 2016.
Appendix A Examples
|Jeopardy!||People of this nation AKA Nippon wrote with a brush, so painting became the preferred form of artistic expression|
|SearchQA||people nation aka nippon wrote brush , painting became preferred form artistic expression|
|Base-NMT||Aka nippon written form artistic expression?|
|AQA-QR||What is name did people nation aka nippon wrote brush expression?|
|AQA-full||people nation aka nippon wrote brush , painting became preferred form artistic expression|
|Jeopardy!||Michael Caine & Steve Martin teamed up as Lawrence & Freddy, a couple of these, the title of a 1988 film|
|SearchQA||michael caine steve martin teamed lawrence freddy , couple , title 1988 film|
|Base-NMT||Who was lawrence of michael caine steve martin?|
|AQA-QR||What is name is name is name michael caine steve martin teamed lawrence freddy and title 1988 film?|
|AQA-full||What is name is name where name is name michael caine steve martin teamed lawrence freddy and title 1988 film key 2000 ?|
|Jeopardy!||Used underwater, ammonia gelatin is a waterproof type of this explosive|
|SearchQA||used underwater , ammonia gelatin waterproof type explosive|
|Base-NMT||Where is ammonia gelatin waterproof?|
|AQA-QR||What is name is used under water with ammonia gelatin water waterproof type explosive?|
|AQA-full||used underwater , ammonia gelatin waterproof type explosive|
|Jeopardy!||The Cleveland Peninsula is about 40 miles northwest of Ketchikan in this state|
|SearchQA||cleveland peninsula 40 miles northwest ketchikan state|
|Base-NMT||The cleveland peninsula 40 miles?|
|AQA-QR||What is name is cleveland peninsula state northwest state state state?|
|AQA-full||What is name are cleveland peninsula state northwest state state state ?|
|Jeopardy!||Tess Ocean, Tinker Bell, Charlotte the Spider|
|SearchQA||tess ocean , tinker bell , charlotte spider|
|Base-NMT||What ocean tess tinker bell?|
|AQA-QR||What is name tess ocean tinker bell link charlotte spider?|
|AQA-full||What is name is name tess ocean tinker bell spider contain charlotte spider contain hump around the world winter au to finish au de mon moist|
|Jeopardy!||During the Tertiary Period, India plowed into Eurasia & this highest mountain range was formed|
|SearchQA||tertiary period , india plowed eurasia highest mountain range formed|
|Bas-NMT||What is eurasia highest mountain range?|
|AQA-QR||What is name were tertiary period in india plowed eurasia?|
|AQA-full||tertiary period , india plowed eurasia highest mountain range formed|
|Jeopardy!||The melody heard here is from the opera about Serse, better known to us as this "X"-rated Persian king|
|SearchQA||melody heard opera serse , better known us x rated persian king|
|Base-NMT||Melody heard opera serse thing?|
|AQA-QR||What is name melody heard opera serse is better persian king?|
|AQA-full||What is name is name melody heard opera serse is better persian king persian K ?|
|Jeopardy!||A type of humorous poem bears the name of this Irish port city|
|SearchQA||type humorous poem bears name irish port city|
|Base-NMT||Name of humorous poem bears name?|
|AQA-QR||What is name is name humorous poem poem bear city city city?|
|AQA-full||What is name is name were humorous poem poem bears name city city city ?|