Papers by Marco Aldo Piccolino-Boniforti
We aim to build a computational model that will help elucidate how humans predict grammatical str... more We aim to build a computational model that will help elucidate how humans predict grammatical structure from acoustic-phonetic detail.
The first step is a proof-of-concept model to distinguish between true and pseudo morphological prefixes in English words, such as discolour, in which dis is a true prefix, and discover, in which dis is a pseudo-prefix. Both words have the same first four phonemes, /dɪsk/ but linguistic and phonetic analyses show that pronunciations of pseudo prefixes tend to have a weaker rhythmic beat than pronunciations of true prefixes have (Ogden et al. 2000; Baker 2008; Baker et al. 2007a) and that these differences affect intelligibility of sentences in noise (Baker 2008; Baker et al. 2007b).
The present work uses Baker‟s original speech corpus and aims to simulate aspects of her observed results. The computational model comprises two main parts. The acoustic signal is first processed within a cochlear model (Patterson et al. 1988; Meddis 1986) that introduces non-linearities in frequency and loudness. The cochlear output is then transformed into an auditory primal sketch (APS, Todd 1994) which simulates perception of amplitude modulation at various temporal resolutions within the auditory system. This representation identifies successive acoustic events in the signal and their so-called relative prominence, a measure that combines amplitude and duration. In the second stage of the present model, the output of the auditory primal sketch is input to a classifier (target class: true vs. pseudo morpheme). Two classifiers are compared, the popular support vector machine (SVM, Vapnik 1995), and the relevance vector machine (RVM, Tipping 2001). The latter seems to display more interesting properties for the simulation of cognitive processes.
The present work reports simulations that compared: 1) RVM vs. SVM; 2) APS based vs. energy based vectors; 3) cochlear-model based versus non-cochlear-model based APS vectors. Model performance was measured both in terms of classification accuracy and model sparsity.
Results show that both RVM and SVM assign the data to the correct true vs pseudo morphological category at well above chance. According to a mixed-effects ANOVA (main factor: RVM vs. SVM; random factor: subject) accuracy difference is just marginally significant. However, the RVM obtains a much sparser representation than the SVM. Comparing APS vs signal energy accuracy using an RVM classifier, energy performs better. Sparsity is not different. All other parameters being equal, and using an RVM classifier, the cochlear model improves the accuracy of the APS compared to the non-cochlear model version. The cochlear model also achieves greater model sparsity.
These results suggest that true prefixes can be reliably distinguished from pseudo prefixes based on the systematic differences in their acoustic patterns, confirming Baker et al.‟s (2007a) findings. Both RVM and the cochlear model show clear advantages in terms of accuracy and/or sparsity. The poorer performance of the auditory primal sketch seems to be linked to the kind of vectors adopted, each one containing a variable number of events.
Workshop Programme, Jan 1, 2008
Proceedings of the …, Jan 1, 2006
As in the previous RTE Challenge, we present a linguistically-based approach for semantic inferen... more As in the previous RTE Challenge, we present a linguistically-based approach for semantic inference which is built around a neat division of labour between two main components: a grammatically-driven subsystem which is responsible for the level of predicate-arguments well-formedness and works on the output of a deep parser that produces augmented head-dependency structures. A second subsystem tries allowed logical and lexical inferences on the basis of different types of structural transformation intended to produce a semantically valid meaning corrispondence. Grammatical relations and semantic roles are used to generate a weighted score. In the current challenge, a number of additional modules have been added to cope with finegrained inferential triggers which were not present in the previous dataset. Different levels of argumenthood have been devised in order to cope with semantic uncertainty generated by nearly-inferrable Text/Hypothesis pairs where the interpretation needs reasoning.
Proc. of ROMAND, Jan 1, 2006
With this selection the workshop unites samples of different techniques for achieving robustness ... more With this selection the workshop unites samples of different techniques for achieving robustness for a range of different processing task. This, however, leaves completely i Wolfgang Menzel ii
Proceedings of the ACL- …, Jan 1, 2007
Recognizing and generating textual entailment and paraphrases are regarded as important technolog... more Recognizing and generating textual entailment and paraphrases are regarded as important technologies in a broad range of NLP applications, including, information extraction, summarization, question answering, information retrieval, machine translation and text generation. Both textual entailment and paraphrasing address relevant aspects of natural language semantics. Entailment is a directional relation between two expressions in which one of them implies the other, whereas paraphrase is a relation in which two expressions convey essentially the same meaning. Indeed, paraphrase can be defined as bi-directional entailment. While it may be debatable how such semantic definitions can be made well-founded, in practice we have already seen evidence that such knowledge is essential for many applications.
Proceedings of the …, Jan 1, 2005
The system for semantic evaluation VENSES (Venice Semantic Evaluation System) is organized as a p... more The system for semantic evaluation VENSES (Venice Semantic Evaluation System) is organized as a pipeline of two subsystems: the first is a reduced version of GETARUN, our system for Text Understanding. The output of the system is a flat list of head-dependent structures (HDS) with Grammatical Relations (GRs) and Semantic Roles (SRs) labels. The evaluation system is made up of two main modules: the first is a sequence of linguistic rule-based subcalls; the second is a quantitatively based measurement of input structures. VENSES measures semantic similarity which may range from identical linguistic items, to synonymous or just morphologically derivable. Both modules go through General Consistency checks which are targeted to high level semantic attributes like presence of modality, negation, and opacity operators, temporal and spatial location checks. Results in cws, accuracy and precision are homogenoues for both training and test corpus and fare higher than 60%.
Uploads
Papers by Marco Aldo Piccolino-Boniforti
The first step is a proof-of-concept model to distinguish between true and pseudo morphological prefixes in English words, such as discolour, in which dis is a true prefix, and discover, in which dis is a pseudo-prefix. Both words have the same first four phonemes, /dɪsk/ but linguistic and phonetic analyses show that pronunciations of pseudo prefixes tend to have a weaker rhythmic beat than pronunciations of true prefixes have (Ogden et al. 2000; Baker 2008; Baker et al. 2007a) and that these differences affect intelligibility of sentences in noise (Baker 2008; Baker et al. 2007b).
The present work uses Baker‟s original speech corpus and aims to simulate aspects of her observed results. The computational model comprises two main parts. The acoustic signal is first processed within a cochlear model (Patterson et al. 1988; Meddis 1986) that introduces non-linearities in frequency and loudness. The cochlear output is then transformed into an auditory primal sketch (APS, Todd 1994) which simulates perception of amplitude modulation at various temporal resolutions within the auditory system. This representation identifies successive acoustic events in the signal and their so-called relative prominence, a measure that combines amplitude and duration. In the second stage of the present model, the output of the auditory primal sketch is input to a classifier (target class: true vs. pseudo morpheme). Two classifiers are compared, the popular support vector machine (SVM, Vapnik 1995), and the relevance vector machine (RVM, Tipping 2001). The latter seems to display more interesting properties for the simulation of cognitive processes.
The present work reports simulations that compared: 1) RVM vs. SVM; 2) APS based vs. energy based vectors; 3) cochlear-model based versus non-cochlear-model based APS vectors. Model performance was measured both in terms of classification accuracy and model sparsity.
Results show that both RVM and SVM assign the data to the correct true vs pseudo morphological category at well above chance. According to a mixed-effects ANOVA (main factor: RVM vs. SVM; random factor: subject) accuracy difference is just marginally significant. However, the RVM obtains a much sparser representation than the SVM. Comparing APS vs signal energy accuracy using an RVM classifier, energy performs better. Sparsity is not different. All other parameters being equal, and using an RVM classifier, the cochlear model improves the accuracy of the APS compared to the non-cochlear model version. The cochlear model also achieves greater model sparsity.
These results suggest that true prefixes can be reliably distinguished from pseudo prefixes based on the systematic differences in their acoustic patterns, confirming Baker et al.‟s (2007a) findings. Both RVM and the cochlear model show clear advantages in terms of accuracy and/or sparsity. The poorer performance of the auditory primal sketch seems to be linked to the kind of vectors adopted, each one containing a variable number of events.
The first step is a proof-of-concept model to distinguish between true and pseudo morphological prefixes in English words, such as discolour, in which dis is a true prefix, and discover, in which dis is a pseudo-prefix. Both words have the same first four phonemes, /dɪsk/ but linguistic and phonetic analyses show that pronunciations of pseudo prefixes tend to have a weaker rhythmic beat than pronunciations of true prefixes have (Ogden et al. 2000; Baker 2008; Baker et al. 2007a) and that these differences affect intelligibility of sentences in noise (Baker 2008; Baker et al. 2007b).
The present work uses Baker‟s original speech corpus and aims to simulate aspects of her observed results. The computational model comprises two main parts. The acoustic signal is first processed within a cochlear model (Patterson et al. 1988; Meddis 1986) that introduces non-linearities in frequency and loudness. The cochlear output is then transformed into an auditory primal sketch (APS, Todd 1994) which simulates perception of amplitude modulation at various temporal resolutions within the auditory system. This representation identifies successive acoustic events in the signal and their so-called relative prominence, a measure that combines amplitude and duration. In the second stage of the present model, the output of the auditory primal sketch is input to a classifier (target class: true vs. pseudo morpheme). Two classifiers are compared, the popular support vector machine (SVM, Vapnik 1995), and the relevance vector machine (RVM, Tipping 2001). The latter seems to display more interesting properties for the simulation of cognitive processes.
The present work reports simulations that compared: 1) RVM vs. SVM; 2) APS based vs. energy based vectors; 3) cochlear-model based versus non-cochlear-model based APS vectors. Model performance was measured both in terms of classification accuracy and model sparsity.
Results show that both RVM and SVM assign the data to the correct true vs pseudo morphological category at well above chance. According to a mixed-effects ANOVA (main factor: RVM vs. SVM; random factor: subject) accuracy difference is just marginally significant. However, the RVM obtains a much sparser representation than the SVM. Comparing APS vs signal energy accuracy using an RVM classifier, energy performs better. Sparsity is not different. All other parameters being equal, and using an RVM classifier, the cochlear model improves the accuracy of the APS compared to the non-cochlear model version. The cochlear model also achieves greater model sparsity.
These results suggest that true prefixes can be reliably distinguished from pseudo prefixes based on the systematic differences in their acoustic patterns, confirming Baker et al.‟s (2007a) findings. Both RVM and the cochlear model show clear advantages in terms of accuracy and/or sparsity. The poorer performance of the auditory primal sketch seems to be linked to the kind of vectors adopted, each one containing a variable number of events.