Cognitive Language Distance Computation

With the increasing demand for life and the increasing cost of learning,language learners hope to learn a language in a relatively short period of time through an efficient way of learning.The theory of cognitive linguistics propose that people learn and use language through cognition and understanding,challenging the traditional theory of linguistics.The popularity of the Internet has brought huge amounts of data,and deep learning technology continues to be heated.This paper will measure the distance between the semantic and the phonetic dimensions from a cognitive perspective,combining deep learning techniques.In people's cognition,the similarity of vocabulary is not only reflected in semantics.For example,the words "father" and "strict",the semantic relevance of these two words is not very strong on the surface,but are closely linked in many people's cognitions.Whether the computational model can be used to simulate the word distance perception model established on the cognitive level is the main content of the cognitive semantic distance computation.In order to achieve this goal,this paper first constructs a cognitive oriented semantic processing dataset from the perspective of people's intuitive association,and then proposes a method of computing semantic distance for cognition,including MLP and CNN based language semantic distance computing.Under the whole union data,the performance of MLP and CNN are better than cosine similarity calculation method.The union strategy of MLP model in different word vector input average F1 value is about 0.7909,CNN model is about 0.7974.The experimental results show that the MLP and CNN-based computational methods proposed in this paper can make up for the gap between word vectors generated from a large number of s tatistical texts and cognitive-oriented semantic distance calculation methods,and they can simulate the distance perception of words in the word association well.In language learning,learners must learn to distinguish similar pronunciation.For easily confused pronunciation such as "bear" and "pair","bell","dear" and other similar pronunciation of these words,it is difficult to identify speech and hearing in people's oral learning.And whether this kind of auditory pronunciation can be simulated by the model is the main content of the cognition-oriented speech distance calculation.Therefore,in this paper,data sets are constructed from people's easily confused pronunciations,then a speech-oriented distance-learning method is proposed,including two-way RNN model based on GRU and LSTM calculation model,CNN calculation model and CNN combined with LSTM calculation model.The results show that on the one edit test set,all methods get the best performance in F1,and CNN+LSTM get the best performance,its F1 is 0.8649.The experimental results show that cognitive-oriented speech distance calculation model can indeed distinguish the speech that is confused easily from the human cognition.