Quantifying Compositionality of Classic and State-of-the-Art Embeddings

• URL: http://arxiv.org/abs/2509.19332v1
• Date: Sun, 14 Sep 2025 18:28:21 GMT
• Title: Quantifying Compositionality of Classic and State-of-the-Art Embeddings
• Abstract summary: We formalize a two-step evaluation that measures the linearity between known entity attributes and their embeddings.<n>We also check reconstruction metrics such as L2 loss, cosine similarity, and retrieval accuracy.<n> Stronger compositional signals are observed in later training stages across data modalities.
• Score: 9.862058044710704
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: For language models to generalize correctly to novel expressions, it is critical that they exploit access compositional meanings when this is justified. Even if we don't know what a "pelp" is, we can use our knowledge of numbers to understand that "ten pelps" makes more pelps than "two pelps". Static word embeddings such as Word2vec made strong, indeed excessive, claims about compositionality. The SOTA generative, transformer models and graph models, however, go too far in the other direction by providing no real limits on shifts in meaning due to context. To quantify the additive compositionality, we formalize a two-step, generalized evaluation that (i) measures the linearity between known entity attributes and their embeddings via canonical correlation analysis, and (ii) evaluates additive generalization by reconstructing embeddings for unseen attribute combinations and checking reconstruction metrics such as L2 loss, cosine similarity, and retrieval accuracy. These metrics also capture failure cases where linear composition breaks down. Sentences, knowledge graphs, and word embeddings are evaluated and tracked the compositionality across all layers and training stages. Stronger compositional signals are observed in later training stages across data modalities, and in deeper layers of the transformer-based model before a decline at the top layer. Code is available at https://github.com/Zhijin-Guo1/quantifying-compositionality.

Related papers

• How do Transformer Embeddings Represent Compositions? A Functional Analysis [5.0310636905746975]
  We test compositionality in Mistral, OpenAI Large, and Google embedding models, and compare them with BERT.<n>We find that ridge regression, albeit linear, best accounts for compositionality.<n>Surprisingly, we find that the classic vector addition model performs almost as well as any other model.
  arXiv  Detail & Related papers  (2025-06-01T09:02:56Z)
• When does compositional structure yield compositional generalization? A kernel theory [0.0]
  We present a theory of compositional generalization in kernel models with fixed, compositionally structured representations.<n>We identify novel failure modes in compositional generalization that arise from biases in the training data.<n>This work examines how statistical structure in the training data can affect compositional generalization.
  arXiv  Detail & Related papers  (2024-05-26T00:50:11Z)
• Learning Syntax Without Planting Trees: Understanding Hierarchical Generalization in Transformers [74.96551626420188]
  Transformers trained on natural language data have been shown to learn its hierarchical structure and generalize to sentences with unseen syntactic structures.<n>We investigate sources of inductive bias in transformer models and their training that could cause such generalization behavior to emerge.
  arXiv  Detail & Related papers  (2024-04-25T07:10:29Z)
• Mutual Exclusivity Training and Primitive Augmentation to Induce Compositionality [84.94877848357896]
  Recent datasets expose the lack of the systematic generalization ability in standard sequence-to-sequence models. We analyze this behavior of seq2seq models and identify two contributing factors: a lack of mutual exclusivity bias and the tendency to memorize whole examples. We show substantial empirical improvements using standard sequence-to-sequence models on two widely-used compositionality datasets.
  arXiv  Detail & Related papers  (2022-11-28T17:36:41Z)
• Pre-trained Sentence Embeddings for Implicit Discourse Relation Classification [26.973476248983477]
  Implicit discourse relations bind smaller linguistic units into coherent texts. We explore the utility of pre-trained sentence embeddings as base representations in a neural network for implicit discourse relation sense classification.
  arXiv  Detail & Related papers  (2022-10-20T04:17:03Z)
• Disentangled Action Recognition with Knowledge Bases [77.77482846456478]
  We aim to improve the generalization ability of the compositional action recognition model to novel verbs or novel nouns. Previous work utilizes verb-noun compositional action nodes in the knowledge graph, making it inefficient to scale. We propose our approach: Disentangled Action Recognition with Knowledge-bases (DARK), which leverages the inherent compositionality of actions.
  arXiv  Detail & Related papers  (2022-07-04T20:19:13Z)
• Inducing Transformer's Compositional Generalization Ability via Auxiliary Sequence Prediction Tasks [86.10875837475783]
  Systematic compositionality is an essential mechanism in human language, allowing the recombination of known parts to create novel expressions. Existing neural models have been shown to lack this basic ability in learning symbolic structures. We propose two auxiliary sequence prediction tasks that track the progress of function and argument semantics.
  arXiv  Detail & Related papers  (2021-09-30T16:41:19Z)
• Distilling Relation Embeddings from Pre-trained Language Models [35.718167335989854]
  We show that it is possible to distill relation embeddings from pre-trained language models. We encode word pairs using a (manually or automatically generated) prompt, and we fine-tune the language model. The resulting relation embeddings are highly competitive on analogy (unsupervised) and relation classification (supervised) benchmarks.
  arXiv  Detail & Related papers  (2021-09-21T15:05:27Z)
• RatE: Relation-Adaptive Translating Embedding for Knowledge Graph Completion [51.64061146389754]
  We propose a relation-adaptive translation function built upon a novel weighted product in complex space. We then present our Relation-adaptive translating Embedding (RatE) approach to score each graph triple.
  arXiv  Detail & Related papers  (2020-10-10T01:30:30Z)
• Structure-Augmented Text Representation Learning for Efficient Knowledge Graph Completion [53.31911669146451]
  Human-curated knowledge graphs provide critical supportive information to various natural language processing tasks. These graphs are usually incomplete, urging auto-completion of them. graph embedding approaches, e.g., TransE, learn structured knowledge via representing graph elements into dense embeddings. textual encoding approaches, e.g., KG-BERT, resort to graph triple's text and triple-level contextualized representations.
  arXiv  Detail & Related papers  (2020-04-30T13:50:34Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.