統計的モデル選択に基づいた連続音声からの語彙学習

Abstract

SummaryThis paper proposes a method for the unsupervised learning of lexicons from pairs of a spoken utterance and an object as its meaning under the condition that any priori linguistic knowledge other than acoustic models of Japanese phonemes is not used. The main problems are the word segmentation of spoken utterances and the learning of the phoneme sequences of the words. To obtain a lexicon, a statistical model, which represents the joint probability of an utterance and an object, is learned based on the minimum description length (MDL) principle. The model consists of three parts: a word list in which each word is represented by a phoneme sequence, a word-bigram model, and a wordmeaning model. Through alternate learning processes of these parts, acoustically, grammatically, and semantically appropriate units of phoneme sequences that cover all utterances are acquired as words. Experimental results show that our model can acquire phoneme sequences of object words with about 83.6% accuracy.

Integrating domain knowledge with user eye gaze in automated word acquisition for conversational interfaces

“…In addition to grounding individual words, previous work has also investigated grounding phrases (referring expressions) to visual objects through semantic decomposition, for example using context free grammars that connect linguistic structures with underlying visual properties [6]. A recent work has applied statistical model selection to automatically acquire lexicons by combining a phoneme acoustic model, a word bigram model, and a word meaning model [20]. Different from above previous work, in our work, the visual perception is indicated by eye gaze.…”

Spatial Concept Acquisition for a Mobile Robot that Integrates Self-Localization and Unsupervised Word Discovery from Spoken Sentences

“…Several studies on language acquisition by robots have assumed that robots have no prior lexical knowledge. These studies differ from speech recognition studies based on a large vocabulary and natural language processing studies based on lexical, syntactic, and semantic knowledge [1], [2]. Studies on Akira Taniguchi and Tadahiro Taniguchi are with Ritsumeikan University, 1-1-1 Noji Higashi, Kusatsu, Shiga 525-8577, Japan (email:a.taniguchi@em.ci.ritsumei.ac.jp; taniguchi@em.ci.ritsumei.ac.jp).…”

Learning a semantic parser from spoken utterances

“…A parser, represented in the form of a lexicon and an inventory containing syntactic constructions, is then estimated based on co-occurrence frequencies, which are often utilized to establish mappings between form -typically words -and meaning (e.g., [7][8] [9]). Alignments are computed both bottom-up by first determining structures of rather low complexity and top-down by including syntactic information; learning linguistic structures of rather low complexity from speech has been addressed previously, e.g., learning (novel) words (e.g., [10][8] [11]) or semantically meaningful sequences, so-called acoustic morphemes (e.g., [12][13] [14]). …”