Still not systematic after all these years: On the compositional skills of sequence-to-sequence recurrent networks

By Brenden M. Lake, Marco Baroni. 2017

83 citations recorded by the source corpus.

Humans can understand and produce new utterances effortlessly, thanks to their systematic compositional skills. Once a person learns the meaning of a new verb "dax," he or she can immediately understand the meaning of "dax twice" or "sing and dax." In this paper, we introduce the SCAN domain, consisting of a set of simple compositional navigation commands paired with the corresponding action sequences. We then test the zero-shot generalization capabilities of a variety of recurrent neural networks (RNNs) trained on SCAN with sequence-to-sequence methods. We find that RNNs can generalize well when the differences between training and test commands are small, so that they can apply "mix-and-match" strategies to solve the task. However, when generalization requires systematic compositional skills (as in the "dax" example above), RNNs fail spectacularly. We conclude with a proof-of-concept experiment in neural machine translation, supporting the conjecture that lack of systematicity is an important factor explaining why neural networks need very large training sets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Model Architectures

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 Computer Vision and Pattern Recognition

In Natural Language Processing

In Model Architectures

Citations in the corpus

The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence (2020, background): “Joel Grus gives a similar example here, with the game fizz-buzz, and Lake and Baroni (Lake & Baroni, 2017) show how some modern natural language systems are vulnerable to similar issues, failing to generalize abstract patterns to novel words in various ways.”
Relational inductive biases, deep learning, and graph networks (2018, background): “…learning about entities and relations (Mitchell, 1980), which we, joining many others (Spelke et al., 1992; Spelke and Kinzler, 2007; Marcus, 2001; Tenenbaum et al., 2011; Lake et al., 2017; Lake and Baroni, 2018; Marcus, 2018b), suggest are an essential ingredient for human-like intelligence.”
Evaluating Compositionality in Sentence Embeddings (2018, background): “The failure of neural network models to achieve compositionality has been a recurring (and controversial) theme in cognitive science (Fodor & Pylyshyn, 1988; Gershman & Tenenbaum, 2015; Lake & Baroni, 2017).”
Measuring abstract reasoning in neural networks (2018, background): “Generalisation in neural networks has been subject of lots of recent debate, with some emphasising the successes (LeCun et al., 2015) and others the failures (Garnelo et al., 2016; Lake & Baroni, 2017; Marcus, 2018).”
Measuring the Diversity of Automatic Image Descriptions (2018, background): “For further discussion of this topic, also see recent work by Lake and Baroni (2017).”
RNNs as psycholinguistic subjects: Syntactic state and grammatical dependency (2018, background): “…al., 2016; Gulordava et al., 2018; Wilcox et al., 2018) and complements a variety of other approaches currently practiced (Shi et al., 2016; Belinkov et al., 2018; Blevins et al., 2018; K´ad´ar et al., 2017; Williams et al., 2018; Et-tinger et al., 2017; Lake and Baroni, 2017; Weber et al., 2018).”
Reasoning with Language Model Prompting: A Survey (2022, background): “Apart from the above-mentioned specific reasoning tasks, there are some benchmarks (Lake and Baroni, 2017; Srivastava et al., 2022) that can evaluate the model’s more diverse and generalized”
Learning to Reason with Third-Order Tensor Products (2018, background): “They are catastrophically affected by systematic [10, 11] differences between training and test data [12, 13, 14, 15].”
The Fine Line between Linguistic Generalization and Failure in Seq2Seq-Attention Models (2018, methodology): “Lake and Baroni (2018) has questioned the systematicity with which these models learn.”
Location Attention for Extrapolation to Longer Sequences (2019, background): “However, many researchers have questioned the forms of generalization that neural networks exhibit, which signiﬁcantly diverges from human-like generalization (Lake and Baroni, 2017; Geirhos et al., 2018).”
A Benchmark for Systematic Generalization in Grounded Language Understanding (2020, methodology): “Since length generalization is entirely unsolved for SCAN, even for methods that made progress on the other splits (Lake & Baroni, 2017; Bastings et al., 2018; Russin et al., 2019; Gordon et al., 2020) , we also separately generate a split to evaluate generalization to longer action sequences.”
Learning compositionally through attentive guidance (2018, background): “One key problem appears to be that neural networks trained with our current learning paradigms do not find solutions exhibit systematic compositionality [6,5], an element which has shown to be crucial for systems to behave in a more humanlike manner [7].”
Memorize or generalize? Searching for a compositional RNN in a haystack (2018, background): “Correspondence to: Adam Liska <adam.liska@gmail.com>. perform well (Lake et al., 2017).”