Recurrent babbling: evaluating the acquisition of grammar from limited input data
This addresses the problem of unrealistic training data in syntax acquisition models for computational linguistics, but it is incremental as it applies an existing method to new data with a novel evaluation approach.
The paper tackled the problem of evaluating grammar acquisition from limited, realistic child-directed input by training an LSTM on a realistically sized subset and analyzing grammatical abstraction in generated output. The result showed that the LSTM abstracts new structures as learning proceeds, though no concrete numbers were provided.
Recurrent Neural Networks (RNNs) have been shown to capture various aspects of syntax from raw linguistic input. In most previous experiments, however, learning happens over unrealistic corpora, which do not reflect the type and amount of data a child would be exposed to. This paper remedies this state of affairs by training a Long Short-Term Memory network (LSTM) over a realistically sized subset of child-directed input. The behaviour of the network is analysed over time using a novel methodology which consists in quantifying the level of grammatical abstraction in the model's generated output (its "babbling"), compared to the language it has been exposed to. We show that the LSTM indeed abstracts new structuresas learning proceeds.