Question 10
A multilingual reporting team pools German and Japanese key phrases into one frequency chart and gets misleading rankings. What aggregation change improves interpretability?
- Aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison
- Use sentence-level spans or offsets when available to align negative classifications with the source text
- Apply a suitable answer confidence threshold and return an approved no-match response
- Match the entity offsets' string-index convention to the client's text-indexing convention
Explanation
Correct Answer
A:Key phrase boundaries and wording differ across languages; cross-language dashboards need deliberate translation or normalization before combining concepts. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after aggregating and normalize extracted key phrases within supported language cohorts before cross-language comparison.
Incorrect Answers
B:That change addresses another failure mode: document-level sentiment does not identify where individual evaluations occur; span metadata supports localized UI highlighting; it does not address this case, for which the relevant intervention is to aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison.
C:Although useful elsewhere in the multilingual customer-support portal, curated question answering can return ranked candidates, but uncertain matches should not be presented as verified policy statements; the present case is different and calls for the team to aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison.
D:NER returns spans relative to an indexing convention; mismatched Unicode indexing, rather than a lack of entity recognition, causes misplaced highlights in its own workflow, but that is a different correction from the present requirement to aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison. The reported issue would remain.
Question 11
A ticket contains both “battery life” and “replacement battery,” which a chart treats as separate top themes. Which post-processing step is appropriate?
- Validate the requested target locale against the Translator language list
- Normalize and group synonymous extracted key phrases for cross-ticket reporting
- Resolve recognized place names against a separate geocoding or gazetteer source
- Apply supported metadata filtering to restrict answers to the employee's division
Explanation
Correct Answer
B:Key-phrase extraction returns text phrases rather than a guaranteed canonical taxonomy; normalization is an application-level reporting choice. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after normalizing and group synonymous extracted key phrases for cross-ticket reporting.
Incorrect Answers
A:Although useful elsewhere in the multilingual customer-support portal, translation targets must be supported language codes; retries or key rotation cannot make an invalid target locale valid; the present case is different and calls for the team to normalize and group synonymous extracted key phrases for cross-ticket reporting.
C:Named entity recognition extracts and categorizes mentions, but it does not independently validate a place name or supply authoritative geographic coordinates in its own workflow, but that is a different correction from the present requirement to normalize and group synonymous extracted key phrases for cross-ticket reporting. The reported issue would remain.
D:That change addresses another failure mode: question-answering metadata can constrain eligible knowledge-base entries before ranking; without scope, a plausible but wrong policy may win; it does not address this case, for which the relevant intervention is to normalize and group synonymous extracted key phrases for cross-ticket reporting.
Question 12
A multilingual application translates technical manuals, but company product names are inconsistently localized. Which change is best for recurring domain terminology?
- Use metadata or application filtering to exclude entries that have expired
- Use an approved terminology customization or supported custom translation workflow
- Apply a confidence-aware review policy instead of relying on the winning sentiment label alone
- Capture entity spans against the source text before applying length-changing redactions
Explanation
Correct Answer
B:Domain terminology requires deliberate control of translated terms; generic translation alone may vary proper names across documents. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after using an approved terminology customization or supported custom translation workflow.
Incorrect Answers
A:That change addresses another failure mode: an answer can match the wording of a question while being no longer valid; date applicability must be enforced by the consuming workflow; it does not address this case, for which the relevant intervention is to use an approved terminology customization or supported custom translation workflow.
C:Although useful elsewhere in the multilingual customer-support portal, sentiment confidence reflects model uncertainty; consistent predicted labels can require different actions when confidence diverges; the present case is different and calls for the team to use an approved terminology customization or supported custom translation workflow.
D:Text replacement alters character positions, so spans extracted after redaction do not necessarily index the unmodified original content in its own workflow, but that is a different correction from the present requirement to use an approved terminology customization or supported custom translation workflow. The reported issue would remain.
Question 13
A privacy workflow substitutes generic placeholders for names before extraction, and the downstream entity offsets no longer match the stored original message. What sequence should be corrected?
- Use entity extraction to verify named-entity categories instead of interpreting key phrases as entities
- Capture entity spans against the source text before applying length-changing redactions
- Use metadata or application filtering to exclude entries that have expired
- Apply a confidence-aware review policy instead of relying on the winning sentiment label alone
Explanation
Correct Answer
B:Text replacement alters character positions, so spans extracted after redaction do not necessarily index the unmodified original content. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after capturing entity spans against the source text before applying length-changing redactions.
Incorrect Answers
A:Although useful elsewhere in the multilingual customer-support portal, a salient phrase need not identify an entity; key-phrase output is not typed person or organization recognition; the present case is different and calls for the team to capture entity spans against the source text before applying length-changing redactions.
C:An answer can match the wording of a question while being no longer valid; date applicability must be enforced by the consuming workflow in its own workflow, but that is a different correction from the present requirement to capture entity spans against the source text before applying length-changing redactions. The reported issue would remain.
D:That change addresses another failure mode: sentiment confidence reflects model uncertainty; consistent predicted labels can require different actions when confidence diverges; it does not address this case, for which the relevant intervention is to capture entity spans against the source text before applying length-changing redactions.
Question 14
A document viewer highlights extracted company names, but the highlights shift when the text contains emoji and surrogate pairs. Which implementation detail should the developer verify before drawing spans?
- Apply a suitable answer confidence threshold and return an approved no-match response
- Use sentence-level spans or offsets when available to align negative classifications with the source text
- Match the entity offsets' string-index convention to the client's text-indexing convention
- Aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison
Explanation
Correct Answer
C:NER returns spans relative to an indexing convention; mismatched Unicode indexing, rather than a lack of entity recognition, causes misplaced highlights. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after matching the entity offsets' string-index convention to the client's text-indexing convention.
Incorrect Answers
A:Curated question answering can return ranked candidates, but uncertain matches should not be presented as verified policy statements in its own workflow, but that is a different correction from the present requirement to match the entity offsets' string-index convention to the client's text-indexing convention. The reported issue would remain.
B:That change addresses another failure mode: document-level sentiment does not identify where individual evaluations occur; span metadata supports localized UI highlighting; it does not address this case, for which the relevant intervention is to match the entity offsets' string-index convention to the client's text-indexing convention.
D:Although useful elsewhere in the multilingual customer-support portal, key phrase boundaries and wording differ across languages; cross-language dashboards need deliberate translation or normalization before combining concepts; the present case is different and calls for the team to match the entity offsets' string-index convention to the client's text-indexing convention.
Question 15
A customer-care UI must underline exactly the sentences marked negative while leaving neutral sentences untouched. What part of sentiment results is needed?
- Aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison
- Set the supported text format to HTML and validate the returned markup
- Apply a suitable answer confidence threshold and return an approved no-match response
- Use sentence-level spans or offsets when available to align negative classifications with the source text
Explanation
Correct Answer
D:Document-level sentiment does not identify where individual evaluations occur; span metadata supports localized UI highlighting. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after using sentence-level spans or offsets when available to align negative classifications with the source text.
Incorrect Answers
A:That change addresses another failure mode: key phrase boundaries and wording differ across languages; cross-language dashboards need deliberate translation or normalization before combining concepts; it does not address this case, for which the relevant intervention is to use sentence-level spans or offsets when available to align negative classifications with the source text.
B:Although useful elsewhere in the multilingual customer-support portal, translator text translation supports an HTML input mode for supported text; treating markup as ordinary plain text risks altering tags; the present case is different and calls for the team to use sentence-level spans or offsets when available to align negative classifications with the source text.
C:Curated question answering can return ranked candidates, but uncertain matches should not be presented as verified policy statements in its own workflow, but that is a different correction from the present requirement to use sentence-level spans or offsets when available to align negative classifications with the source text. The reported issue would remain.
Question 16
A support application receives a 400 response when requesting translation into an unsupported locale code. What should be checked first?
- Resolve recognized place names against a separate geocoding or gazetteer source
- Apply supported metadata filtering to restrict answers to the employee's division
- Normalize and group synonymous extracted key phrases for cross-ticket reporting
- Validate the requested target locale against the Translator language list
Explanation
Correct Answer
D:Translation targets must be supported language codes; retries or key rotation cannot make an invalid target locale valid. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after validating the requested target locale against the Translator language list.
Incorrect Answers
A:Although useful elsewhere in the multilingual customer-support portal, named entity recognition extracts and categorizes mentions, but it does not independently validate a place name or supply authoritative geographic coordinates; the present case is different and calls for the team to validate the requested target locale against the Translator language list.
B:That change addresses another failure mode: question-answering metadata can constrain eligible knowledge-base entries before ranking; without scope, a plausible but wrong policy may win; it does not address this case, for which the relevant intervention is to validate the requested target locale against the Translator language list.
C:Key-phrase extraction returns text phrases rather than a guaranteed canonical taxonomy; normalization is an application-level reporting choice in its own workflow, but that is a different correction from the present requirement to validate the requested target locale against the Translator language list. The reported issue would remain.
Question 17
Support tickets contain the word 'Apple', which may refer to a fruit or a technology company. The application must link a mention to the appropriate well-known knowledge-base entity when one is supported, rather than only label its general type. Which capability should be evaluated?
- Use language sentiment confidence as the canonical identifier of the mentioned entity
- Evaluate contextual entity linking and use its returned linked entity identifiers when available
- Treat every occurrence of 'Apple' as the organization category returned by named entity recognition
- Use key-phrase extraction because any repeated phrase automatically maps to one encyclopedia entry
Explanation
Correct Answer
B:Entity linking uses surrounding context to disambiguate a mention and associate it with a recognized knowledge-base entity. This differs from named entity recognition, which primarily identifies mentions and categories rather than guaranteeing a canonical external identity.
Incorrect Answers
A:Sentiment confidence describes polarity assessment rather than an entity identity or knowledge-base link. It cannot distinguish a fruit mention from a corporation reference in the required way.
C:A category such as Organization does not prove that every appearance refers to the company. The requirement is a disambiguated linked entity, so categorizing the surface text alone can produce incorrect identities.
D:Key-phrase extraction finds salient text spans; it does not necessarily resolve ambiguous wording to a canonical entity. A recurring phrase such as 'Apple' still requires contextual disambiguation and supported linking behavior.
Question 18
A custom FAQ bot invents answers when none of the curated entries matches an unfamiliar question. What response policy would limit this?
- Apply a suitable answer confidence threshold and return an approved no-match response
- Use sentence-level spans or offsets when available to align negative classifications with the source text
- Aggregate and normalize extracted key phrases within supported language cohorts before cross-language comparison
- Match the entity offsets' string-index convention to the client's text-indexing convention
Explanation
Correct Answer
A:Curated question answering can return ranked candidates, but uncertain matches should not be presented as verified policy statements. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after applying a suitable answer confidence threshold and return an approved no-match response.
Incorrect Answers
B:Document-level sentiment does not identify where individual evaluations occur; span metadata supports localized UI highlighting in its own workflow, but that is a different correction from the present requirement to apply a suitable answer confidence threshold and return an approved no-match response. The reported issue would remain.
C:That change addresses another failure mode: key phrase boundaries and wording differ across languages; cross-language dashboards need deliberate translation or normalization before combining concepts; it does not address this case, for which the relevant intervention is to apply a suitable answer confidence threshold and return an approved no-match response.
D:Although useful elsewhere in the multilingual customer-support portal, nER returns spans relative to an indexing convention; mismatched Unicode indexing, rather than a lack of entity recognition, causes misplaced highlights; the present case is different and calls for the team to apply a suitable answer confidence threshold and return an approved no-match response.
Question 19
The system automatically escalates every message not classified as positive, including neutral questions. Which change corrects the decision logic?
- Apply supported metadata filtering to restrict answers to the employee's division
- Handle neutral and mixed sentiment explicitly and consider the returned confidence values
- Resolve recognized place names against a separate geocoding or gazetteer source
- Normalize and group synonymous extracted key phrases for cross-ticket reporting
Explanation
Correct Answer
B:Sentiment results are not merely positive versus negative; treating neutral feedback as a complaint creates avoidable false escalations. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after handling neutral and mixed sentiment explicitly and consider the returned confidence values.
Incorrect Answers
A:That change addresses another failure mode: question-answering metadata can constrain eligible knowledge-base entries before ranking; without scope, a plausible but wrong policy may win; it does not address this case, for which the relevant intervention is to handle neutral and mixed sentiment explicitly and consider the returned confidence values.
C:Named entity recognition extracts and categorizes mentions, but it does not independently validate a place name or supply authoritative geographic coordinates in its own workflow, but that is a different correction from the present requirement to handle neutral and mixed sentiment explicitly and consider the returned confidence values. The reported issue would remain.
D:Although useful elsewhere in the multilingual customer-support portal, key-phrase extraction returns text phrases rather than a guaranteed canonical taxonomy; normalization is an application-level reporting choice; the present case is different and calls for the team to handle neutral and mixed sentiment explicitly and consider the returned confidence values.
Question 20
A product team treats every extracted phrase as a confirmed brand or person and sends an alert. What correction is required?
- Apply a confidence-aware review policy instead of relying on the winning sentiment label alone
- Capture entity spans against the source text before applying length-changing redactions
- Use entity extraction to verify named-entity categories instead of interpreting key phrases as entities
- Use an approved terminology customization or supported custom translation workflow
Explanation
Correct Answer
C:A salient phrase need not identify an entity; key-phrase output is not typed person or organization recognition. In the multilingual customer-support portal, confirm the fix by retesting the affected behavior after using entity extraction to verify named-entity categories instead of interpreting key phrases as entities.
Incorrect Answers
A:Sentiment confidence reflects model uncertainty; consistent predicted labels can require different actions when confidence diverges in its own workflow, but that is a different correction from the present requirement to use entity extraction to verify named-entity categories instead of interpreting key phrases as entities. The reported issue would remain.
B:That change addresses another failure mode: text replacement alters character positions, so spans extracted after redaction do not necessarily index the unmodified original content; it does not address this case, for which the relevant intervention is to use entity extraction to verify named-entity categories instead of interpreting key phrases as entities.
D:Although useful elsewhere in the multilingual customer-support portal, domain terminology requires deliberate control of translated terms; generic translation alone may vary proper names across documents; the present case is different and calls for the team to use entity extraction to verify named-entity categories instead of interpreting key phrases as entities.