direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Part-of-Speech Tagging wtih Neural Networks for a conversational agent

LOCATION:  TEL, Auditorium 3 (20th floor), Ernst-Reuter-Platz 7, 10587 Berlin

Date/Time: 18.06.2018, 15:00-15:45   

SPEAKER: Andreas Müller (TU Berlin)


A part-of-speech tagger is a system which automatically assigns the part of speech to words using contextual information. Potential applications for part- of-speech taggers exist in many areas of computational linguistics including speech recognition, speech synthesis, machine translation or information retrieval in general.
The part-of-speech tagging task of natural language processing is also used in the advisory artificial conversational agent called ALEX. ALEX was developed to answer questions about modules and courses at the Technische Universität Berlin. The system takes the written natural language requests from the user and tries to transform them into SQL-queries. To understand the natural language queries, the system uses a Hidden Markov Model (HMM) to assign tags to each word of the query (part-of-speech tagging). This HMM tagger is trained with manually created training templates that are filled with the data in the database to be queried. The manually created sentence-templates and the slot-filling resulted in many training data sentences with the same structure. This often led to wrong tagging results when the HMM tagger was presented with an input sentence, having a structure that doesn’t occur in the training templates.
This thesis shows two different neural network approaches for the language modeling of the input sentences and evaluates and compares both neural net- work based tagger as well as the HMM based tagger.

Zusatzinformationen / Extras


Schnellnavigation zur Seite über Nummerneingabe