RTJTN: Relational Triplet Joint Tagging Network for Joint Entity and Relation Extraction

Extracting entities and relations from unstructured sentences is one of the most concerned tasks in the field of natural language processing. However, most existing works process entity and relation information in a certain order and suffer from the error iteration. In this paper, we introduce a relational triplet joint tagging network (RTJTN), which is divided into joint entities and relations tagging layer and relational triplet judgment layer. In the joint tagging layer, instead of extracting entity and relation separately, we propose a tagging method that allows the model to simultaneously extract entities and relations in unstructured sentences to prevent the error iteration; and, in order to solve the relation overlapping problem, we propose a relational triplet judgment network to judge the correct triples among the group of triples with the same relation in a sentence. In the experiment, we evaluate our network on the English public dataset NYT and the Chinese public datasets DuIE 2.0 and CMED. The F1 score of our model is improved by 1.1, 6.0, and 5.1 compared to the best baseline model on NYT, DuIE 2.0, and CMED datasets, respectively. In-depth analysis of the model’s performance on overlapping problems and sentence complexity problems shows that our model has different gains in all cases.

Syntactic complexity in translated and non-translated texts: A corpus-based study of simplification

This study approaches the investigation of the simplification hypotheses in corpus-based translation studies from a syntactic complexity perspective. The research is based on two comparable corpora, the English monolingual part of COCE (Corpus of Chinese-English) and the native English corpus of FLOB (Freiburg-LOB Corpus of British English). Using the 13 syntactic complexity measures falling into five subconstructs (i.e. length of production unit, amount of subordination, amount of coordination, phrasal complexity and overall sentence complexity), our results show that translation as a whole is less complex compared to non-translation, reflected most prominently in the amount of subordination and overall sentence complexity. Further pairwise comparison of the four subgenres of the corpora shows mixed results. Specifically, the translated news is homogenous to native news as evidenced by the complexity measures; the translated genres of general prose and academic writing are less complex compared to their native counterparts while translated fiction is more complex than non-translated fiction. It was found that mean sentence length always produced a significant effect on syntactic complexity, with higher syntactic complexity for longer sentence lengths in both corpora. ANOVA test shows a highly significant main effect of translation status, with higher syntactic complexity in the non-translated texts (FLOB) than the translated texts (COCE), which provides support for the simplification hypothesis in translation. It is also found that, apart from translation status, genre is an important variable in affecting the complexity level of translated texts. Our study offers new insights into the investigation of simplification hypothesis from the perspective of translation from English into Chinese.

Dependency Parsing-based Entity Relation Extraction over Chinese Complex Text

Open Relation Extraction (ORE) plays a significant role in the field of Information Extraction. It breaks the limitation that traditional relation extraction must pre-define relational types in the annotated corpus and specific domains restrictions, to realize the goal of extracting entities and the relation between entities in the open domain. However, with the increase of sentence complexity, the precision and recall of Entity Relation Extraction will be significantly reduced. To solve this problem, we present an unsupervised Clause_CORE method based on Chinese grammar and dependency parsing features. Clause_CORE is used for complex sentences processing, including decomposing complex sentence and dynamically complementing sentence components, which can reduce sentences complexity and maintain the integrity of sentences at the same time. Then, we perform dependency parsing for complete sentences and implement open entity relation extraction based on the model constructed by Chinese grammar rules. The experimental results show that the performance of Clause_CORE method is better than that of other advanced Chinese ORE systems on Wikipedia and Sina news datasets, which proves the correctness and effectiveness of the method. The results on mixed datasets of news data and encyclopedia data prove the generalization and portability of the method.

Comparing predictors of sentence self-paced reading times: Syntactic complexity versus transitional probability metrics

When estimating the influence of sentence complexity on reading, researchers typically opt for one of two main approaches: Measuring syntactic complexity (SC) or transitional probability (TP). Comparisons of the predictive power of both approaches have yielded mixed results. To address this inconsistency, we conducted a self-paced reading experiment. Participants read sentences of varying syntactic complexity. From two alternatives, we selected the set of SC and TP measures, respectively, that provided the best fit to the self-paced reading data. We then compared the contributions of the SC and TP measures to reading times when entered into the same model. Our results showed that both measures explained significant portions of variance in self-paced reading times. Thus, researchers aiming to measure sentence complexity should take both SC and TP into account. All of the analyses were conducted with and without control variables known to influence reading times (word/sentence length, word frequency and word position) to showcase how the effects of SC and TP change in the presence of the control variables.

When Hearing Does Not Mean Understanding: On the Neural Processing of Syntactically Complex Sentences by Listeners With Hearing Loss

Purpose Adults with mild-to-moderate age-related hearing loss typically exhibit issues with speech understanding, but their processing of syntactically complex sentences is not well understood. We test the hypothesis that listeners with hearing loss' difficulties with comprehension and processing of syntactically complex sentences are due to the processing of degraded input interfering with the successful processing of complex sentences. Method We performed a neuroimaging study with a sentence comprehension task, varying sentence complexity (through subject–object order and verb–arguments order) and cognitive demands (presence or absence of a secondary task) within subjects. Groups of older subjects with hearing loss ( n = 20) and age-matched normal-hearing controls ( n = 20) were tested. Results The comprehension data show effects of syntactic complexity and hearing ability, with normal-hearing controls outperforming listeners with hearing loss, seemingly more so on syntactically complex sentences. The secondary task did not influence off-line comprehension. The imaging data show effects of group, sentence complexity, and task, with listeners with hearing loss showing decreased activation in typical speech processing areas, such as the inferior frontal gyrus and superior temporal gyrus. No interactions between group, sentence complexity, and task were found in the neuroimaging data. Conclusions The results suggest that listeners with hearing loss process speech differently from their normal-hearing peers, possibly due to the increased demands of processing degraded auditory input. Increased cognitive demands by means of a secondary visual shape processing task influence neural sentence processing, but no evidence was found that it does so in a different way for listeners with hearing loss and normal-hearing listeners.