From Language Models over Tokens to Language Models over Characters

Modern language models are internally -- and mathematically -- distributions over token strings rather than \emph{character} strings, posing numerous challenges for programmers building user applications on top of them. For example, if a prompt is specified as a character string, it must be tokenized before passing it to the token-level language model. Thus, the tokenizer and consequent analyses are very sensitive to the specification of the prompt (e.g., if the prompt ends with a space or not). This paper presents algorithms for converting token-level language models to character-level ones. We present both exact and approximate algorithms. In the empirical portion of the paper, we benchmark the practical runtime and approximation quality. We find that -- even with a small computation budget -- our method is able to accurately approximate the character-level distribution (less than 0.00021 excess bits / character) at reasonably fast speeds (46.3 characters / second) on the Llama 3.1 8B language model.

Reframing linguistic bootstrapping as joint inference using visually-grounded grammar induction models

Semantic and syntactic bootstrapping posit that children use their prior knowledge of one linguistic domain, say syntactic relations, to help later acquire another, such as the meanings of new words. Empirical results supporting both theories may tempt us to believe that these are different learning strategies, where one may precede the other. Here, we argue that they are instead both contingent on a more general learning strategy for language acquisition: joint learning. Using a series of neural visually-grounded grammar induction models, we demonstrate that both syntactic and semantic bootstrapping effects are strongest when syntax and semantics are learnt simultaneously. Joint learning results in better grammar induction, realistic lexical category learning, and better interpretations of novel sentence and verb meanings. Joint learning makes language acquisition easier for learners by mutually constraining the hypotheses spaces for both syntax and semantics. Studying the dynamics of joint inference over many input sources and modalities represents an important new direction for language modeling and learning research in both cognitive sciences and AI, as it may help us explain how language can be acquired in more constrained learning settings.

Correlation Does Not Imply Compensation: Complexity and Irregularity in the Lexicon

It has been claimed that within a language, morphologically irregular words are more likely to be phonotactically simple and morphologically regular words are more likely to be phonotactically complex. This inverse correlation has been demonstrated in English for a small sample of words, but has yet to be shown for a larger sample of languages. Furthermore, frequency and word length are known to influence both phonotactic complexity and morphological irregularity, and they may be confounding factors in this relationship. Therefore, we examine the relationships between all pairs of these four variables both to assess the robustness of previous findings using improved methodology and as a step towards understanding the underlying causal relationship. Using information-theoretic measures of phonotactic complexity and morphological irregularity (Pimentel et al., 2020; Wu et al., 2019) on 25 languages from UniMorph, we find that there is evidence of a positive relationship between morphological irregularity and phonotactic complexity within languages on average, although the direction varies within individual languages. We also find weak evidence of a negative relationship between word length and morphological irregularity that had not been previously identified, and that some existing findings about the relationships between these four variables are not as robust as previously thought.

The Stable Entropy Hypothesis and Entropy-Aware Decoding: An Analysis and Algorithm for Robust Natural Language Generation

State-of-the-art language generation models can degenerate when applied to open-ended generation problems such as text completion, story generation, or dialog modeling. This degeneration usually shows up in the form of incoherence, lack of vocabulary diversity, and self-repetition or copying from the context. In this paper, we postulate that ``human-like'' generations usually lie in a narrow and nearly flat entropy band, and violation of these entropy bounds correlates with degenerate behavior. Our experiments show that this stable narrow entropy zone exists across models, tasks, and domains and confirm the hypothesis that violations of this zone correlate with degeneration. We then use this insight to propose an entropy-aware decoding algorithm that respects these entropy bounds resulting in less degenerate, more contextual, and "human-like" language generation in open-ended text generation settings.

Evaluating Distributional Distortion in Neural Language Modeling

A fundamental characteristic of natural language is the high rate at which speakers produce novel expressions. Because of this novelty, a heavy-tail of rare events accounts for a significant amount of the total probability mass of distributions in language (Baayen, 2001). Standard language modeling metrics such as perplexity quantify the performance of language models (LM) in aggregate. As a result, we have relatively little understanding of whether neural LMs accurately estimate the probability of sequences in this heavy-tail of rare events. To address this gap, we develop a controlled evaluation scheme which uses generative models trained on natural data as artificial languages from which we can exactly compute sequence probabilities. Training LMs on generations from these artificial languages, we compare the sequence-level probability estimates given by LMs to the true probabilities in the target language. Our experiments reveal that LSTM and Transformer language models (i) systematically underestimate the probability of sequences drawn from the target language, and (ii) do so more severely for less-probable sequences. Investigating where this probability mass went, (iii) we find that LMs tend to overestimate the probability of ill formed (perturbed) sequences. In addition, we find that this underestimation behaviour (iv) is weakened, but not eliminated by greater amounts of training data, and (v) is exacerbated for target distributions with lower entropy.

Systematic Generalization with Edge Transformers

Recent research suggests that systematic generalization in natural language understanding remains a challenge for state-of-the-art neural models such as Transformers and Graph Neural Networks. To tackle this challenge, we propose Edge Transformer, a new model that combines inspiration from Transformers and rule-based symbolic AI. The first key idea in Edge Transformers is to associate vector states with every edge, that is, with every pair of input nodes -- as opposed to just every node, as it is done in the Transformer model. The second major innovation is a triangular attention mechanism that updates edge representations in a way that is inspired by unification from logic programming. We evaluate Edge Transformer on compositional generalization benchmarks in relational reasoning, semantic parsing, and dependency parsing. In all three settings, the Edge Transformer outperforms Relation-aware, Universal and classical Transformer baselines.

Compositional Generalization in Dependency Parsing

Compositionality, or the ability to combine familiar units like words into novel phrases and sentences, has been the focus of intense interest in artificial intelligence in recent years. To test compositional generalization in semantic parsing, Keysers et al. (2020) introduced Compositional Freebase Queries (CFQ). This dataset maximizes the similarity between the test and train distributions over primitive units, like words, while maximizing the compound divergence: the dissimilarity between test and train distributions over larger structures, like phrases. Dependency parsing, however, lacks a compositional generalization benchmark. In this work, we introduce a gold-standard set of dependency parses for CFQ, and use this to analyze the behavior of a state-of-the art dependency parser (Qi et al., 2020) on the CFQ dataset. We find that increasing compound divergence degrades dependency parsing performance, although not as dramatically as semantic parsing performance. Additionally, we find the performance of the dependency parser does not uniformly degrade relative to compound divergence, and the parser performs differently on different splits with the same compound divergence. We explore a number of hypotheses for what causes the non-uniform degradation in dependency parsing performance, and identify a number of syntactic structures that drive the dependency parser's lower performance on the most challenging splits.

Linguistic dependencies and statistical dependence

What is the relationship between linguistic dependencies and statistical dependence? Building on earlier work in NLP and cognitive science, we study this question. We introduce a contextualized version of pointwise mutual information (CPMI), using pretrained language models to estimate probabilities of words in context. Extracting dependency trees which maximize CPMI, we compare the resulting structures against gold dependencies. Overall, we find that these maximum-CPMI trees correspond to linguistic dependencies more often than trees extracted from non-contextual PMI estimate, but only roughly as often as a simple baseline formed by connecting adjacent words. We also provide evidence that the extent to which the two kinds of dependency align cannot be explained by the distance between words or by the category of the dependency relation. Finally, our analysis sheds some light on the differences between large pretrained language models, specifically in the kinds of inductive biases they encode.

Characterizing Idioms: Conventionality and Contingency

Idioms are unlike other phrases in two important ways. First, the words in an idiom have unconventional meanings. Second, the unconventional meaning of words in an idiom are contingent on the presence of the other words in the idiom. Linguistic theories disagree about whether these two properties depend on one another, as well as whether special theoretical machinery is needed to accommodate idioms. We define two measures that correspond to these two properties, and we show that idioms fall at the expected intersection of the two dimensions, but that the dimensions themselves are not correlated. Our results suggest that idioms are no more anomalous than other types of phrases, and that introducing special machinery to handle idioms may not be warranted.

Jointly Learning Truth-Conditional Denotations and Groundings using Parallel Attention

We present a model that jointly learns the denotations of words together with their groundings using a truth-conditional semantics. Our model builds on the neurosymbolic approach of Mao et al. (2019), learning to ground objects in the CLEVR dataset (Johnson et al., 2017) using a novel parallel attention mechanism. The model achieves state of the art performance on visual question answering, learning to detect and ground objects with question performance as the only training signal. We also show that the model is able to learn flexible non-canonical groundings just by adjusting answers to questions in the training set.