Gábor András Recski

Competence in lexical semantics

We investigate from the competence standpoint two recent models of lexical semantics, algebraic conceptual representations and continuous vector models.

Measuring Semantic Similarity of Words Using Concept Networks

We present a state-of-the-art algorithm for measuring the semantic similarity of word pairs using novel combinations of word embeddings, WordNet, and the concept dictionary 4lang. We evaluate our system on the SimLex-999 benchmark data. Our top score of 0.76 is higher than any published system that we are aware of, well beyond the average inter-annotator agreement of 0.67, and close to the 0.78 average correlation between a human rater and the average of all other ratings, suggesting that our system has achieved nearhuman performance on this benchmark.

MathLingBudapest: Concept Networks for Semantic Similarity

We present our approach to measuring semantic similarity of sentence pairs used in Semeval 2015 tasks 1 and 2. We adopt the sentence alignment framework of (Han et al., 2013) and experiment with several measures of word similarity. We hybridize the common vector-based models with definition graphs from the 4lang concept dictionary and devise a measure of graph similarity that yields good results on training data. We did not address the specific challenges posed by Twitter data, and this is reflected in placing 11th from 30 in Task 1, but our systems perform fairly well on the generic datasets of Task 2, with the hybrid approach placing 11th among 78 runs.

Hungarian Noun Phrase Extraction Using Rule-based and Hybrid Methods

We implement and revise Kornais grammar of Hungarian NPs [11] to create a parser that identifies noun phrases in Hungarian text. After making several practical amendments to our morphological annotation system of choice, we proceed to formulate rules to account for some specific phenomena of the Hungarian language not covered by the original rule system. Although the performance of the final parser is still inferior to state-of-the-art machine learning methods, we use its output successfully to improve the performance of one such system.

Computational methods in semantics

Meaning Representation, or AMR (Banarescu et al., 2013), is a more recent formalism for representing the meaning of linguistic structures as directed graphs. The last few years have seen a rise in AMR-related work, including a corpus of AMR-annotated text (Banarescu et al., 2013), several approaches to generating AMRs from running text (Vanderwende et al., 2015; Peng et al., 2015; Pust et al., 2015), and various applications to computational semantics (Pan et al., 2015; Liu et al., 2015). Figure 2.8: AMR representation of The boy wants to go (Banarescu et al., 2013, p.179) Nodes of AMR graphs represent concepts of two basic types: they are either English words, or framesets from PropBank (Palmer et al., 2005), used to abstract away from English syntax. PropBank framesets are essentially English verbs (or verb-particle con-