Scribd
Upload
9 views

LING1003A Tutorial 6

Linguistics tutorial

Uploaded by

sirisha.moonsamy
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
9 views

LING1003A Tutorial 6

Linguistics tutorial

Uploaded by

sirisha.moonsamy
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

LING1003A: Tutorial 6

Sirisha Moonsamy
Student number: 2682988
Tutorial Group 1

1. Describe three aspects of what a computer has to do in order to use language. (3)

 When interacting with language, a computer must analyse the incoming text and
comprehend its structure and meaning. This entails breaking phrases down into
grammatical components like nouns, verbs, and adjectives and identifying how
these components connect to one another. Understanding the text's context and
semantics is critical for correctly interpreting the user's purpose.

 A computer must understand the meaning intricacies of the language in addition


to the grammatical structure. This includes recognising synonyms, antonyms, and
the many figurative uses of words. To deliver meaningful responses, the
computer must be able to discern inferred meanings, context, and allusions to
real-world entities or concepts.

 A computer's effective use of language requires not only processing input but
also providing logical and contextually relevant responses. This necessitates the
computer selecting acceptable words, phrases, and sentence structures to
accurately deliver its message. The response should consider the input, the
history of the conversation, and the desired tone or style of communication.

2. What problems did early researchers face while developing speech recognition software?
(6)

 Speaker Variability: Due to differences in pitch, tone, accent, and speech


patterns, computers struggled to understand speech from diverse speakers. The
software struggled to adapt to these differences, resulting in poor recognition
performance when confronted with speakers with varying vocal characteristics.

 Out-of-Vocabulary Words: In general, speech recognition software was created


with a pre-set vocabulary in mind. If a user said a term that wasn't in this
vocabulary, the system wouldn't recognise it. Because of this constraint, the
software was unable to interpret or reply to words that were not on its predefined
list.

 Run-Together Words: The software had trouble identifying between individual


words when words were pronounced in quick succession or run together in
casual speech. As a result, the system's identification accuracy suffered as it
failed to partition and identify the individual words inside a sentence. Incorrect
Pronunciation: If words were not properly uttered or well enunciated, speech
recognition software would frequently fail to recognise them correctly. Incomplete
pronunciation, muttering, or slurring of words made correct recognition difficult.
 Effects of Coarticulation: Coarticulation is the phenomenon in which the
pronunciation of one phoneme is modified by the phonemes before or following it
in a sequence. This generated issues since the software's word representation
may not account for coarticulation effects.

 Phonetic Similarity: The vocabulary terms had to be chosen with phonetic


similarity in mind. If the vocabulary comprised terms that were phonetically
similar, the software would struggle to distinguish between them, resulting in
more errors and confusion.

3. Explain how the cocktail party effect causes problems for a speech recogniser. (2)

 In real-world circumstances, numerous speakers are frequently speaking at the


same time, as are diverse background noises. Humans can use the cocktail party
effect to selectively listen to one speaker while dismissing others. However, the
existence of several overlapping sound sources makes it challenging for a
speech recognizer to effectively detect and isolate individual speech signals. As a
result, there is a lot of noise and interference, which makes it difficult for the
system to identify and comprehend specific speech patterns.

 The cocktail party effect is based on the ability of the brain to segment and
recognise speech depending on spatial clues, linguistic context, and speaker
characteristics. In contrast, speech recognition algorithms lack the fundamental
cognitive mechanisms that enable this level of spatial and contextual awareness.
As a result, the system may struggle to appropriately partition spoken words and
assign them to their individual speakers, resulting in confusion and inaccuracies
while transcribing or interpreting the information.

a. Define a lexical decision task and a naming task. (4)

 Lexical Decision Task: A lexical decision task is a cognitive experiment used in


psychology and linguistics to assess people's ability to determine whether a
presented stimulus is a true word or a non-word. Participants in this activity are
often shown a sequence of letter strings that are either actual words (for
example, "apple") or non-words (for example, "blort"). Their duty is to swiftly
determine whether or not each string constitutes a valid word by pressing a
specific key. The amount of time it takes to make these decisions and the
accuracy with which they respond provide insights into the brain processes
involved in word recognition and lexical access.

 Naming task: A naming task is a strategy used to research language processing


and cognitive functions related to word retrieval and creation. Participants in this
task are shown visual stimuli, often words or pictures, and are instructed to
verbally produce the name of the presented item as rapidly as feasible. The time
it takes to say the stimulus's name is measured and analysed. Naming tasks are
frequently used in research to study aspects such as word frequency, semantic
priming, and the impact of context on word retrieval and speech output.
b. What are the measures of interest in both tasks? (2)

Measures of lexical decision task interest:

 Response Time: The amount of time it takes for participants to decide


whether a presented stimulus is a word or not. Faster response times could
imply faster word recognition and lexical access.
 Accuracy: The proportion of correct responses in determining whether or not
a stimulus is a real word. Higher accuracy indicates a greater capacity to
distinguish between words and non-words.

Measures of interest in the naming task:

 Naming Latency: The amount of time it takes for participants to name a particular
stimulus. Faster naming latencies imply efficient word retrieval and output.
 Naming Accuracy: The proportion of correctly named stimuli produced. Higher
accuracy suggests better language creation and word recall abilities.

4. Explain the two approaches used in speech synthesis. (4)

Concatenative Synthesis: This method entails generating a database of pre-recorded speech


units (phonemes, diphones, or tiny chunks of speech) spoken by a real person. During
synthesis, these units are concatenated and aligned to produce the required speech output.
This technology can produce highly lifelike and expressive speech, but it requires a vast
amount of recorded data and can be difficult to adjust or adapt.

Formants: Formants are discrete frequency bands that contribute to the quality and timbre of
speech sounds. Formant synthesis produces speech by varying the frequencies of these
formants to produce various phonemes and words. This method allows for more control over
the qualities of the synthesised speech, but it may lack the naturalness found in
concatenative synthesis.

5. Why do Garden path sentences require the listener to shift their analysis midway through
the sentence? (4)

Garden path sentences require the listener to re-evaluate the sentence's structure or meaning
halfway through the sentence because they initially lead the listener to parse the sentence in a
way that appears grammatically correct, but then introduce unexpected elements that force a re-
evaluation of the sentence's structure or meaning. This shift in analysis occurs as a result of
ambiguity or various viable interpretations of the initial words, resulting in a "garden path" effect
in which the listener's thinking is misguided before they realise the correct interpretation.
For example, "the old man the boats. "The word "man" is originally regarded as the subject of the
sentence in this statement, leading the listener to expect a verb to follow it. However, as the
sentence develops, it becomes evident that "man" is the subject of the word "old," and "old man"
is the object of the verb "old." The correct reading necessitates a shift in comprehension halfway
through the statement. This phenomena demonstrates how our parsing of phrases is influenced
by our grammatical expectations, resulting in brief confusion before the proper meaning is
determined.

You might also like