Natural Language Speech Processing
The most difficult problem in AI is to process natural language by computers or elsewhere. In other words, natural language processing is the most difficult problem in artificial intelligence. If we are talking about the major problems of NLP, then one of the major problems of NLP is speech processing - building theories and models of how utterances stick together to form a coherent speech / b>. In fact, language still consists of co-located, structured, and coherent groups of sentences rather than isolated, unrelated sentences like movies. These cohesive groups of sentences are called speeches.
Concept of coherence
The coherence and structure of discourse are interconnected in several ways. Consistency, along with the property of good text, is used to assess the output quality of the snatural language generation system. The question that arises here is: what does it mean for a text to be consistent? Suppose we have collected a sentence from every page of the newspaper, then will it be a speech? Of course not. This is because these sentences are inconsistent. The coherent speech must have the following properties -
Relation of coherence between the statements
The speech would be coherent if it had significant connections between its statements. This property is called a consistency relation. For example, some kind of explanation must be there to justify the link between the statements.
Relation between entities
Another property which makes a discourse coherent is that there must be some type of relation with the entities. This type of consistency is called entity-based consistency.
Structure of discourse
An important question regarding discourse is what type of structure the speech must have. The answer to this question depends on the segmentation we applied to the speech. Speech segmentations can be defined as determining the types of structures for a great speech. Speech segmentation is quite difficult to implement, but it is very important for info search, text summary, and info retrieval applications.
Speech segmentation algorithms
In this section, we will discover the speech segmentation algorithms. The algorithms are described below -
Unsupervised Speech Segmentation
The class of unsupervised speech segmentation is often represented as linear segmentation. We can understand the task of linear segmentation with the help of an example. In the example, there is a task of segmenting the text into multi-paragraph units; unitedtees represent the passage from the original text. These algorithms depend on the cohesion which can be defined as the use of certain linguistic devices to link the textual units together. On the other hand, lexicon cohesion is the cohesion which is indicated by the relationship between two or more words in two units such as the use of synonyms.
Segmentation of supervised speech
The previous method has no hand-labeled segment boundaries. On the other hand, supervised segmentation of speech must have training data labeled by limits. It is very easy to acquire the same. In supervised segmentation of speech, speech markers or cue words play an important role. The speech marker or cue word is a word or phrase that serves to indicate the structure of speech. These speech markers are domain specific.
The repetitionLexical tion is a way to find structure in a speech, but it does not satisfy the requirement of being a coherent speech. To achieve coherent discourse, one must focus on coherence relations in particular. As we know, the coherence relation defines the possible connection between the utterances in a speech. Hebb proposed this type of relationship as follows -
We take two terms S 0 and S 1 to represent the meaning of the two linked sentences -
It deduces that the state asserted by the term S 0 could cause the state asserted by S1 . For example, two statements show the result of the relationship: Ram was caught in the fire. His skin burned.
It deduces that the state claimed by S1 could cause the state claimed by S 0 . For example, two declarations show the relationship - Ram got into a fight with Shyam 's friend. He was drunk.
It deduces p (a1, a2, ...) from the assertion of S0 and p ( b1, b2,…) from the S1 assertion. Here ai and bi are similar for all i. For example, two statements are parallel - Ram wanted a car. Shyam wanted money.
It deduces the same proposition P from the two assertions - S0 and S 1 For example, two statements show the development of the relationship: Ram was from Chandigarh. Shyam was from Kerala.
This occurs when a change of state can be inferred from the assertion of S 0 , whose final state can be deduced from S1 and vice versa. For example, both statements show the occasion of the relationship: Ram took the book. He gave it to Shyam.
Building a hierarchical structuree of speech
The coherence of all the speech can also be considered by the hierarchical structure between the coherence relations. For example, the following passage can be represented as a hierarchical structure -
S 1 - Ram went to the bank to deposit some money.
S 2 - He then took a train to Shyam's clothing store.
S3 - He wanted to buy clothes.
S 4 - There is no new clothes for the party.
S 5 - He also wanted to talk to Shyam reg arding about his health
Interpreting the sentences of any speech is another important task and to achieve it we need to know who or what entity is spokensure. Here, the interpretation reference is key. The reference can be defined as the linguistic expression designating an entity or an individual. For example, in the passage, Ram , the manager of ABC Bank , saw his friend Shyam in a store. He went to meet him, linguistic expressions like Ram, His, He refer.
On the same note, reference resolution can be defined as the task of determining which entities are referenced by which linguistic expression.
Terminology used in the reference resolution
We use the following terminologies in the reference resolution -
Reference expression - The natural language expression that is used to make a reference is called a reference expression. For example, the passage used above is a reference expression.
Referrer - This is the entity that is referenced. For example, in the last example given, Ram is a referrer.
Corefer - When two expressions are used to refer to the same entity, they are called corefers. For example, Ram and he are corefers.
Background - The term is licensed to use another term. For example, Ram is the antecedent for the reference he.
Anaphora & Anaphoric - It can be defined as the reference to an entity that was previously introduced in the sentence. And, the referent expression is called anaphoric.
Discourse model - The model that contains representations of entities that have been ref erred in speech and the relationship in which they are engaged.
Types of referent expressions
Now let's see the different typesreferring expressions. The five types of reference expressions are described below -
This type of reference represents entities that are new to the listener in the context of speech. For example - in the sentence Ram had once gone to bring him food - some are an undefined reference.
Defined noun sentences
Opposite above, this type of reference represents entities that are not new or identifiable to the listener in the context of the speech. For example, in the sentence - I used to read the Times of India - the Times of India is a definitive reference.
This is a definitive form of reference. For example, Ram laughed as hard as he could. The word he represents the pronoun reference expression.
These demonstrate and behave differently from simple defined pronounses. For example, this and that are demonstrative pronouns.
This is the simplest type of reference expression. It can also be the name of a person, organization, and location. For example, in the examples above, Ram is the name arbitration expression.
Benchmark resolution tasks
The two benchmark resolution tasks are described below.
This is about finding reference expressions in text that refer to the same entity. Simply put, it 's the task of finding corefer expressions. A set of coreferring expressions is called a coreference string. For example - He, Chief Manager and His - these are reference expressions in the first passage given as an example.
Constraint on coreference resolution
In English, the main problem for coreference resolution is the pronoun il. The reason behind this is that the pronoun he has many uses. For example, he can refer a lot like him and her. The pronoun it also refers to things that do not refer to specific things. For example, it's raining. It's really good.
Solving the pronominal anaphora
Unlike coreference solving, solving the pronominal anaphora can be defined as the task of finding the antecedent of 'a single pronoun. For example, the pronoun is his and the task of solving the pronominal anaphora is to find the word Ram because Ram is the antecedent.