The Natural Language Processing In Siri

Natural language is the tool to represent Information. It is the ability for the users to communicate with any system or device in a conversational manner without any conversational hindrances.

Natural Language Processing in Siri

The above topic constitutes the working of Siri. This topic now being covered is how Siri can able to process sentences and convert them into meaningful texts. Natural Language Processing has become more advanced to understand human voice and convert them into text and manipulating those texts and giving us useful answers. Apple has introduced Dictation in Siri, a Speech to text feature. It process natural language, contextual and conversational and works with built in apps on Siri like calendars, contacts address etc.

Function of the application Siri

The application Siri requires a speech input. Using Automatic Speech Recognition System it translates our voice to text. This system analyses an individual’s voice and fine tunes the voice to get more accurate result in the form of commands and questions1. The voice recognition does not just try to understand voice and translates to text but it tries to understand who the speaker or the person is. This will then try to learn the person’s speech and Siri trains according to it. Next, the texts are sent to servers on the Apple Siri cloud and these words uses the Natural Language processing store to execute a series of understanding language models of the text. Uses POS tagging(Part of Speech), noun phrase chunking, dependency and parsing to parse text to form a meaningful sentences. Using various models like Boolean, vector and language changes a lot more and giving a meaning to the text. These meanings are taken from the lexical group of corpus documents finding the probability of the next words sequences. The higher the probability of the words the more chances the words get joined till the sentences gets structured. These parsed texts use the question and intent analysis to analyze the parsed text. For eg. “Call my mother”. It has to first store in its local memory who is your mother. When it is stored, the next time it will know who you are mentioning about it. Another example if you ask Siri to “Schedule a meeting”. Siri has to use pragmatic approach to understand what exactly the user is trying to describe. From the word “Schedule” it has to find out the whether the schedule is related to time table or a calendar. But still Siri can understand with one piece of word. It has to analyze the next piece of sentences to match a meaning. But since a Siri has a database for understanding this small information, it can actually make out that the calendar needs to be opened and ask the users to input the words and time of the meeting. If the answer can’t be found or can’t directly answer it, Apple Siri then uses the data mash up technology combining two or more third party web services to get information out of the questions asked by the user. These third party mash up like Open Table, Wolfram Alpha perform actions, operations and question and answering. These answers are produced back into natural language processed text from a given question. And then uses the TTS technology to transform the Natural Language processed text to synthesized speech.

Speech Recognition to Natural Language Processing

Apple Siri uses speech synthesizer to process the voice to be sent to a server to transcribe the words spoken into parsed text. The sound spoken is encoded into compact digital form that preserves its information. The signal from the connected phone is relayed wireless through a nearby Cell tower then to the ISP which is then communicated to a server in the cloud which is been loaded with a variety of speech models to understand the language. The Speech is also verified local on the Apple Siri Devices if it can manage what the user is spoken in small measure otherwise recognizer which is installed locally communicates it to the server recognizer to evaluate the speech. The Server is then compared with a speech against a statistical model to estimate the sounds spoken and the order in which it was spoken.

Techniques in Natural Language processing

They are no of techniques used for constructing a good natural sentence such as statistical modeling, lexical and grammatical parsing, machine learning. These technologies deconstruct words, sentences, paragraphs and entire documents expressed in human language and map them to a semantic structure. 3

If you ask a question to Siri what is the time in New Delhi, India? This information is simplified by mapping out spoken words to correct spellings. Once the text is correctly categorized, it will extract information from various sources semantically.

In this case, we have two elements to be retrieved, the time and the location as New Delhi.

Siri must know what it is relates to. For example, 100 Degree of Celsius, the application must understand that at 100 Degrees it is hot. So there is a requirement for a Domain Knowledge.

Definition of Techniques of Natural Language processing working with Apple Siri

Language models are assigned a probability to sequence of words. It captures the properties of the language and predicts the next words in the sequence. If the Language model is used in information retrieval, it takes information from documents and takes the probability of higher number of words used for the next sentences. 6. Since apple Siri is a closed propriety, we assume these language models are used to constructs these words. They probably might use different methods to form sentences when asked to Siri. Some of the definitions of the certain techniques could have been used in Siri to process natural language are just given below

Statistical modeling – This algorithm are models which corrects words and sentences. It resolves difficulties of ambiguous data which are processed with realistic grammar. 4

Semantic Searching -It uses entity extraction which takes nouns, places and people and maps all these into a single concept.

Parsing – It is the process of analyzing a text made of sequence of tokens to determine grammatical structure to construct the words in a grammatical form.5

Auto Categorization – It auto categorize thousands and thousands of words and then sort out of the works according to the words spoken and very quickly.

Machine Translation – This concept automatically tries to translate one text from one form to another. It has to know grammar, semantics and facts about the real world in order to make accurate translation.

Sentiment Analysis – It checks the tone of an article whether it is positive, negative or neutral. It extracts the subjects of the information.